View
3
Download
0
Category
Preview:
Citation preview
Higher Education Supply, Local Competition and
Economic Performance: Evidence from Italy∗
Elena Cottini (Universita’ Cattolica di Milano),
Paolo Ghinetti (Universita’ del Piemonte Orientale)
Simone Moriconi (Ieseg School of Management)
This draft: January 2018
Abstract
This paper exploits an own built dataset on the history of higher education institutions in
Italy during 1861-2010, to analyse local competition effects in the supply of higher education.
We measure local supply of higher education as the number of teaching units (faculties) at the
province level, and analyse how local supply responds to the supply of neighbouring provinces.
Our identification strategy relies on instrumental variables that exploit initial conditions of the
Italian higher education supply, inherited by the states that existed before the Italian unifi-
cation. We find that local competition effects are sizeable: the more conservative estimates
show that having 8 more faculties in the neighbourhood reduces the local supply by about 1
faculty. Such competition forces are mostly concentrated within the same field of study, the
same region, and a spatial reach of 90 Km. Finally, we estimate the economic returns of higher
education supply, and show that the opportunity cost of competition forces may be sizeable,
in terms of foregone per capita value added. However, this cost is compensated by positive
economic externalities coming from the higher education supply in the neighborhood.
Key Words: local competition; higher education supply; historical data; initial conditions.
JEL codes: I23, I28, N00, R1
∗Address: Elena Cottini, Department of Economics and Finance, Universita’ Cattolica di Milano,
Largo Gemelli 1, 20123 Milano (MI), Italy; elena.cottini@unicatt.it; Paolo Ghinetti, Dipartimento di Studi
per l’Economia e l’Impresa, Universita’ del Piemonte Orientale, Via Perrone 18, Novara (NO), Italy:
paolo.ghinetti@uniupo.it; Simone Moriconi, Department of Economics and Quantitative Methods, Ieseg
School of Management and LEM, 1 parvis de La Defense - 92044 Paris - La Defense Cedex (France);
s.moriconi@ieseg.fr.
1
1 Introduction
Throughout the past century, OECD countries experienced a major expansion of their higher
education (HE) systems, both in terms of students enrolled and number of institutions
involved in the market. This process, particularly during the last few decades was the
result of a local diffusion and differentiation of HE supply (e.g. in terms of autonomy,
territorial attachment, etc. See Eurydice, 2008), aiming at featuring HE systems which could
match the increasing development needs at the regional and sub-regional (e.g. provincial)
level (OECD, 1999). As a result of this process, HE providers nowadays operate in more
competitive environments, which explains different performances of HE institutions (Aghion
et al., 2010), as well as cross-regional differentials in regional productivity and income per
capita (Gennaioli et al., 2013; Valero and Van Reenen, 2016).
The present paper contributes to this debate by investigating competition in the provision
of HE services at the local level. HE providers are nowadays regarded as a major source of
local development. They may attract private investments, and enhance local human capital.
HE institutions, particularly in highly centralised states, provide key local agencies that
bring together within the territory national interests in science and technology, industrial
performance, education and skills, health, social inclusion and culture (OECD, 2008). For
these reasons, the modern approach to the provision of HE is a bottom-up one, which
moves from the needs of regional or even provincial development: the establishment of a
HE provider in a province, may greatly benefit the local stakeholders, by diverting resources
there, perhaps at the expense of neighbouring provinces which see reduced opportunities to
open their own HE institution.
The primary goal of this paper is to analyse the existence of local competition in the
supply of HE, by estimating the response of HE supply in an Italian province, to the supply
of neighbours. Our analysis accounts for the fact that local HE supply may have feedback
effects on HE supply of neighbours, or they can be both determined by external factors. To
address such reverse causality and endogeneity issues, we exploit exogenous variation coming
from the geographical distribution of HE supply inherited by pre-unitarian Italian states (i.e.
the initial conditions of the Italian HE supply), interacted with state-level reforms of the
Italian university system. Italy is indeed the country with the longest history of university
education in the Western world. The first Studium was opened in Bologna, and dates back
to 1088; many other institutions were established by the XII century.1 We exploit the
1In the Middle-Age a Studium was an autonomous and secular organisation among students that choseand personally funded teachers through a donation collection system (collectio). In most cases, a Studiumanticipated the foundation of a University (Pini, 2000). Besides Bologna, by the beginning of the XII
2
expansion of pre-unitarian HE supply triggered by national HE reforms during the Italian
history to identify competition forces. We analyse whether competition forces operate within
or across broad fields of studies and discuss the boundaries of competition effects. In the
final part of the paper, we build upon the recent literature that looks at the economic effect
of HE, to evaluate the economic returns of HE supply in terms of value added per capita.
This allows us to measure the opportunity cost (in terms of foregone value added per capita)
of HE competition, and compare this cost with the direct economic spillovers coming from
HE supply of neighbours.
To these purposes, we constructed the History of Italian Universities (HIU) dataset, a
cross-province panel which follows over time 110 Italian provinces between 1861, which is
the year of Italian unification, and 2010, when a major reform radically changed the Italian
university system. For each province, we record the local supply of HE institutions, measured
as the number of faculties - the university teaching unit responsible for all BSc or MSc courses
within a given field of study - registered within each province in a given year (in total, and
by area of science). We then match the cross-province panel dataset to its mirror images
and register pairwise relationship between province couples based either on a contiguity, or
a distance criterion, or both to estimate competition effects, following the main approaches
by the economics literature (see e.g. Davies and Vadlamannati, 2013; Parchet, 2014).
A desirable feature of our data is the long time span (150 years), which goes back at
the onset of Italian unification, which provides substantial variability in the supply of HE at
the province level. We control for local unobserved heterogeneity by including province and
region-by-year fixed effects. Our data also include province level indicators (i.e. total and ac-
tive population, size of the population cohort aged 0-14, relative size of agricultural/industry
and services sectors), which provide useful control variables throughout the analysis. We can
identify the competition effects by an instrumental variables procedure that exploits exoge-
nous variation associated with the initial conditions of Italian higher education interacted
with HE reforms that occurred during Italian history. This empirical strategy relies on
two identifying assumptions. The first is that the cross-province distribution of the supply
of higher education at the onset of Italian unification is exogenous. This “pre-unitarian”
HE supply was indeed inherited by the states that existed before the Italian unification
i.e. determined by exogenous factors such as culture, politics, institutions, and geography
(Squicciarini and Voigtlander, 2015). The second identifying assumption is that single Italian
provinces due not have enough “voice” (due to their weak political relevance) to influence
century, on the Italian territory there were active Studia in Modena (1175), Napoli (1224), Perugia (1308),Siena (1240), and Roma (1303). In Parma the first Studium established in 962, was closed in the XIIcentury and then re-opened in 1601 (See Brizzi and Romano, 2007 for details).
3
state-level decisions. By taking the interactions of state-level reforms with pre-unitarian
provinces, we assume that each pre-unitarian HE system is affected by national reforms, but
the intensity of treatment varies across systems.
We find evidence that HE reforms deeply shaped supply patterns in provinces where
HE institutions would already be present at the onset of Italian history. The reform that
liberalised access to university education in 1969 is particularly relevant, as it created strong
incentives to expand local HE supply. Our main empirical results point to strong competition
effects at the province level: our more conservative estimates suggest that having 8 faculties
more in the neighbourhood reduces local HE supply by over 1 faculty. This is a non-negligible
impact, considering that the expansion of higher education that took place in the second half
of the 20th century led to an increase by 4 faculties per province, on average. We show that
local displacement forces mostly operate within the same field of study, the same region, and
a 90 Km linear distance. Finally, our estimates show that the economic returns of one more
faculty in a province is 0.8%−1.4% of local value added. This suggests that the opportunity
cost of competition forces may be sizeable, in economic terms. However, our estimates also
point out that this cost is over-compensated by direct positive externalities coming from HE
supply in the neighbourhood.
The rest of the paper is organised as follows. In the next section we provide background
information about the Italian HE system, as well as we frame our analysis in the literature,
from both a theoretical and empirical point of view. In Section 3 we describe the data used
in the analysis. The empirical analysis and main results of competition in HE supply are
discussed in Section 4. Section 5 discusses the opportunity cost of competition effects in
terms of economic value of HE supply. Section 6 concludes.
2 Background and literature
2.1 Institutional and historical setting
The Italian provision of higher education is the prototypical European one. Figure 1 doc-
uments the evolution of HE supply in France, Italy, Germany and UK. Starting from the
mid-nineteenth century, all four countries are characterised by a common pattern i.e. a posi-
tive and moderate increasing trend in the supply of higher education, with a marked increase
in the pace of expansion starting from the 1960s.2 In all four countries, HE is regulated by
2Madsen and Murtin (2017) find also a very similar trend in education attainments for the UK. Theirtime series starts much earlier than ours i.e. in the XIIIth century, and refers to total years of schooling (i.e.
4
Figure 1: Evolution in the supply of higher education 1859-2009 in France (FR), Germany(DE), UK and Italy (IT): n. universities
Notes: authors calculations on HIU and WHED UNESCO data.
the central government, which put in place reforms that favoured the expansion of higher
education (Eurydice, 2008). While it possesses a flatter profile relative to the other three
countries, the growth of HE supply in Italy was substantial, and in 2009 the number of
universities was 3 times larger than in 1859.3
In line with European standards, HE providers in Italy are highly differentiated between
old and new ones, large and small, public and private ones. The faculty is the relevant HE
institution. This is the teaching unit, and it is meant to govern the supply of HE in a given
field of study. The genesis of a faculty is very often detached from that of a university and
often tailored to respond to local demand of HE services. In the history of the Italian HE
system there are many cases in which the creation of the faculty is antecedent that of the
university. In pre-war period, this was due to institutional constraints as very few disciplines
were formally taught within the university framework.4 In the post-war period, HE supply
in a province was often established through the creation of new faculties. Their genesis
was often supported by province-level consortia, which gathered all competences needed to
includes primary-to-tertiary education).3There are several reasons for the Italian trend, for example the persistently lower number of high school
graduates compared to countries like Germany, especially in the last 40 years.4At the onset of Italian history there were only medicine, law, humanities, mathematics and natural
sciences. All the scientific knowledge and the social sciences were taught by schools (equivalent to singlefaculty institutions) recognised by the state and part of the HE system. They granted degrees in professionaland technical subjects equivalent to university degrees on a legal standpoint, but were not formally consideredas university education, until legislative reforms in 1923 and 1933. As an example, the Veterinary School ofMilan was already there at the moment of Italian unification, while the University of Milan was establishedin 1923. The School was formally annexed to the university in 1924 and transformed in Faculty in 1934 (seeon-line appendix).
5
the project: a HE provider (i.e. an existing university, not necessarily located in the same
province, and willing to expand its market and differentiate its supply), financing institutions
and economic stakeholders (generally banks, chambers of commerce or local investors), and
local politicians (particularly at the province-level), enjoying direct connections with central
government’s representatives. Each consortium would present a faculty start-up project,
which would then be evaluated by the central state.5
Traditionally, HE providers enjoy some degree of local autonomy, with strong attachment
to the territory and its economic and political stakeholders (OECD, 2008). This has historical
reasons, reflecting the fact that the Italian HE system at the time of the unification was
the simple aggregation of the HE systems of the ancient pre-unitarian Italian states. The
province has always been the relevant units for the provision of higher education services.
On a political ground, this is the oldest governance level on the Italian territory, equivalent
to French departments and to US counties, which was inherited from pre-unitarian Italian
States. From a territorial perspective, these are the intermediate units between the region
and the municipality, and the level where one can observe the local demand of HE and
the local effects of human capital on industrialisation patterns and structure of production
(Squicciarini and Voigtlander, 2015; Ciccone and Peri, 2011), or wages (Ciccone and Peri,
2006; Bratti and Leombruni, 2014).
Table 1 presents the HE supply of the Italian Kingdom inherited by pre-unitarian States.
In 1870, by the end of the main wave of Italian unification (i.e. after Papal States were
annexed to the Kingdom of Italy), there were more than 80 faculties already operating. This
pre-unitarian supply was geographically dispersed in 21 out of the 69 provinces on the Italian
territory at that time.6 The capital(s) of each pre-unitarian states (e.g. Palermo and Napoli
in the Kingdom of the Two Sicilies), had their own HE institutions. Capitals of old duchies,
which used to be independent states during the Middle-Age, and annexed to a pre-unitarian
state later on (e.g. the Duchies of Ferrara and Pesaro Urbino, annexed by the Papal States in
5In the majority of cases, the presentation of project proposal would follow specific calls and developmentplans by the central state itself. Following this procedure, for example, in 1993 seven new faculties wereopened in the Provinces of Novara, Vercelli, and Alessandria, as separate branches of the University ofTorino. These gained autonomy in 1998, by the creation of the brand-new University of Eastern Piedmont.This case is not isolated, and several universities have faculties in multiple provinces (see on-line appendixfor details).
6At the moment of unification, the first Italian government reorganised provincial constituencies takinginto account most of the maps that historically different states had built over time (Palombelli, 2012). Thisadministrative reorganisation implied the transformation of several pre-unitarian provinces into districtsthat were annexed to existing provinces to obtain larger territorial and administrative units. Many districtsbecame autonomous again later on during Italian history, so that the number of provinces in our sampleincreases from 69 in 1870 to 110 in 2010. In the empirical analysis we account for this process, and checkthe robustness of our results to alternative definitions of pre-unitarian provinces.
6
the XVI-XVII centuries) had their own HE institutions too. Each institution complied with
the accreditation rules of its own state only. The diversity of pre-unitarian HE system would
also derive from cultural fragmentation: in 1861 only the 2.5% of the Italian population
would be able to speak Italian, while the rest of the population would only use their local
regional language (De Mauro, 1970).
The heterogeneity of pre-unitarian HE systems emerged clearly from the assessment of
HE institutions present on the Italian territory by the Law 3725/1859 of the Kingdom of
Sardinia (called Casati, from the name of the Minister of Education).7 This assessment
introduced a clear ranking in the Italian university system, which is described in Column
[5] of Table 1: only in 9 out of the 21 pre-unitarian provinces with an active institution,
HE supply fully complied with the declared quality standard of the Kingdom of Sardinia
(A-level supply in Column [5]). In several provinces, HE supply would not fully match the
quality standards, but provided second-tier regional higher education (B-level in Column [5]).
In two provinces (Sassari, and Macerata), local HE institutions fell short of the minimum
requirements to be considered higher education (C-level in Column [5]). The Casati Law
also defined the requirements for private universities, which were not financed by the state.
Private provision was a typical feature of the HE systems in the old provinces of Ferrara,
Urbino and Perugia, where universities were financed by the Duchies and the Vatican (Brizzi
and Romano, 2007).
Italian governments before WWII put a lot of effort to harmonise pre-unitarian HE
systems and create a homogeneous national HE system. Two main regulatory reforms were
implemented during this period. The Law 2102/1923 (called Gentile, from the name of the
Minister of Education), introduced new disciplines (e.g. Political Sciences), granted academic
status to technical studies, and ensures adequate financial resources to the HE system.8 The
reform also restored the distinction between A-level and B-level HE institutions, which had
been progressively excised over the years (see on-line appendix for details). The reform
launched by the Law 1592/1933 (called De Vecchi-Bottai) centralised higher education, by
nationalising private universities and recognising technical and applied schools as faculties
with full academic status.
Post-WWII governments put in place reforms that produced a new institutional setting
7This law set out the rules for accrediting pre-existing institutions into the new Italian university system.It was initially applied to new territories of the Kingdom of Sardinia. The successive Matteucci Law in 1862extended it to all territories that gradually entered the Italian Kingdom.
8The reform expands the perimeter of the higher education system to include specialised schools intechnical discipline such as engineering and architecture, economics, management, commercial and socialsciences. Before the reform, these Schools were HE institutions legally recognised by the State, but haddifferent rules (access criteria, financing, validity of degrees, etc) and stood outside the University system.
7
Table 1: Pre-unitarian HE supply
[1] Pre-unitarian [2] Pre-unitarian [3] HE year of [4] No. of [5] HE assessmentProvince State appearance Faculties Casati LawBologna Papal States 1088 6 ACagliari Kin. of Sardinia 1620 4 BCatania Kin. of two Sicilies 1445 4 BFerrara Papal States 1391 4 privateGenova Kin. of Sardinia 1481 5 B
Macerata Papal States 1540 3 CMessina Kin. of two Sicilies 1838 3 BMilano Lombardy-Venetia 1791 2 A,AModena Duchy of Modena 1175 4 BNapoli Kin. of two Sicilies 1224 7 A
Palermo Kin. of two Sicilies 1806 4 APadova Lombardy-Venetia 1407 4 APerugia Papal States 1308 2 private
Pisa Gran Duchy of Tuscany 1343 6 AParma Duchy of Parma 962 5 B
P.Urbino Papal States 1671 2 privatePavia Kin. of Sardinia 1361 5 ARoma Papal States 1303 5 ASiena Gran Duchy of Tuscany 1240 2 B
Sassari Kin. of Sardinia 1765 3 CTorino Kin. of Sardinia 1404 8 A,B
Note: In Column [3], HE year of appearance refers to the year of the first studium in theprovince. In Milano HE supply appears with the autonomous School of Veterinary Studies.Column [5] reports the quality assessment of HE institution(s) in the province according to theevaluation framework set by the Casati Law. This evaluation refers to public institutions only.HE provision was private in Ferrara, Perugia, and P.Urbino. In Torino, the Casati Law assignsA-score to the University of Torino, and a B score to the Polytechnic. In Milano, it assignsA-score to both the Scientific Academy, and the Polytechnic (see on-line appendix for details).Source is History of Italian Universities (HIU) Data.
of the national university system. This was previously organised to serve the “elite”, later on
was asked to provide higher education for the “mass”. The Law 910/1969 liberalised higher
education access to all students with a 5 year diploma of secondary education, including
those from technical schools (that before 1969 did not allow university enrolment). The
consequent rise in enrolment rates put a lot of pressure on the national university system to
adapt HE supply to the massive increase in demand. Indeed, the Law 766/1973 legislates the
opening of new faculties and increases the number of faculty hires in public universities. The
Law 382/1980 reorganised the internal governance of universities as well as the recruitment
and career of university professors. Successive Laws 392/1989 and 245-341/1990 reorganise
the system of allocation of funding to higher education.9 The Law 59/1997 (called Bassanini
9The Law establishes a dedicated “Ministry of Universities and Scientific and Technological Research”,and sets up triennial development plans for universities (see on-line appendix for details).
8
Figure 2: Evolution of HE supply in Italy in 1860-2010.
Notes: authors calculations on HIU data.
reform) granted universities more financial and teaching autonomy.
This transition from national elite to mass higher education took place during the same
period everywhere in Europe (Eurydice, 1999). This is consistent with evidence presented
in Figure 1 above for France, Germany, and the UK. A similar process also regarded the US
(Smith, 2010). As detailed in Figure 2, the rise in HE supply after 1969, involved Humanities,
STEM, and Social Sciences in a roughly similar way, notwithstanding a constantly lower HE
supply in the former discipline relative to the latter two.
2.2 Literature review
The micro-foundations of our study are consistent with a monopolistically competitive loca-
tion model featuring endogenous choices of students and HE institutions. Students consume
higher education services and display preferences for a differentiated consumption (say a HE
degree in a given field of study, or by a specific institution), and have commuting costs to
the location of the HE institution. In this class of models, the spatial equilibrium depends
on the trade-off between students’ utility from consuming their preferred HE variety and
minimising commuting costs. The spatial equilibrium features a dispersed HE supply, which
is designed both to “retain” local students with high commuting costs, and attract students
from close locations, by supplying their preferred HE variety.10
10At the equilibrium, the coordination costs of a spatially decentralised system more than compensatethe advantages of having more varieties in the supply of HE at the local level. Accordingly, a decentralisedequlibrium features too much spatial differentiation as compared to the social optimum in Hotelling or Saloplocation models, so that a centralised solution would be preferable from a welfare standpoint.
9
Figure 3: Students’ mobility and choice of HE institution.
Notes: authors’ calculations on data from the Italian survey of high school graduates, 2007 (ISTAT).
This dispersion is consistent with such a spatial equilibrium under the assumption that
students have low mobility (i.e. high commuting cost). Figure 3 reports the distribution of
high school graduates in Italy by distance from the university they enrolled to. It suggests
that students’ mobility is very low: the majority of Italian high school graduates chose a
HE institution at a distance of about 100 Km. Mobility is higher among students with high
aptitudes towards HE and it is particularly low among low aptitude students.11 This recalls
well known results for the US (see Hoxby, 2016).12
Our paper relates to the literature that studies the long-run changes of education institu-
tions. Woesmann (2003) and Schtz et al. (2005) use student-level data from 39 countries to
trace back international differences in students performance to key cross-country differences
in institutional design (e.g. with respect to the degree of school autonomy, competition from
private schools, and extent of equality in the education opportunities). Braga et al. (2013)
exploit variation associated with reforms to education institutions in 24 countries over a time
span of 70 years, to identify the causal effect of institutions on educational attainments. The
long time span allows the authors to control for time invariant, country-specific factors.
They show that education reforms that expand access to education increase average years of
schooling and reduce educational inequality. Reforms that foster school accountability and
11We define students’ aptitude based on their final grade at lower secondary schools. High aptitudestudents are those with the highest grade (4), while low aptitute students are those with the lowest grades(1 or 2), where grades are measured on a four point scale.
12Hoxby (2016) shows that the probability that a low aptitude students chooses a college at more than 250miles of distance is less than 10% in 2010. The probability is equal to 50% for an average aptitude studentand 90% for a high aptitude student.
10
autonomy, are also found to increase average years of schooling, but also increase inequal-
ity.13 We contribute to this literature in several respects. The aforementioned studies relate
institutional differences to educational attainments whereas we look at the input of the ed-
ucational process. Also, our paper focuses on post-compulsory education institutions while
existing studies look at either compulsory education only (Woesmann, 2003; Schtz et al.,
2005), or both compulsory and post-compulsory education (Braga et al., 2013). Finally, the
studies mentioned above use data from several countries, while we use historical data for
one specific country, Italy. This allows us to avoid cross-country unobserved heterogeneity
and exploit exogenous variation associated with the initial conditions of the Italian HE to
investigate the effect of HE reforms occurring during Italian history on HE supply. The
main objective of our analysis is to exploit this source of exogenous variation to make causal
inference on the degree of local competition in the provision of HE services, which is a novel
topic in the economics’ literature.
The final part of the paper, where we estimate the economic returns of higher education
supply, relates to the literature that looks at the long-term economic effects of education
institutions. A first strand of this literature, inspired by Unified Growth Theories (Galor,
2005), adopts an historical perspective and discusses the role of initial conditions in the stock
of human capital as a crucial element for industrialisation e.g. in France, Prussia, and UK
(see respectively Squicciarini and Voigtlander, 2015; Becker et al., 2011; Madsen and Murtin,
2017).14 In particular, our identification strategy is similar to Squicciarini and Voigtlander
(2015), which exploit the exogenous cross-department distribution of subscriptions to the
Encyclopedie in France to identify the impact of “upper-tail knowledge” on city growth
after the onset of French industrialisation. The second strand analyses the economic returns
of education institutions, also in the presence of spatial externalities (Gennaioli et al., 2013;
Valero and Van Reenen, 2016; Ertur and Koch, 2007). Our empirical analysis is closer to
Valero and Van Reenen (2016). They use a novel dataset, with panel information on the
establishment, and regional location of nearly 15000 universities across 78 countries for the
13The literature is much more vast when it comes to specific policy and reforms, which however do notentail an institutional change. For example, Hoxby and Turner (2014) evaluate policy interventions thatimprove the access of low-income students to college related information (i.e. application process, college netcosts). They show that this intervention makes low-income high performing students more likely to applyand being admitted to more colleges, especially those with higher graduation rates and offering studentsmore instructional amenities.
14Becker et al. (2011) use the education level observed before the start of industrialisation (i.e. in 1816)as an instrument to identify the effect of education on industrialisation in Prussia. Similarly, Madsen andMurtin (2017) use 80 and 190 years lagged values of years of schooling to identify the impact of schoolingon real GDP growth in the UK during 1270-2010. They show an important contribution of human capitalaccumulation (years of schooling) before and after the Industrial Revolution. This view is not undisputed.See for example, Clark (2005) for a discussion of the (lack of a) role of human capital accumulation for theindustrial revolution in the UK.
11
post-war period. They estimate fixed effects models at the sub-national level and find that
a doubling of universities in a region is associated with about 4 percent higher future GDP
per capita. Our analysis differs as it concentrates on Italy, covering its 150 years history.
It allows for observations at the NUTS3 level, measuring HE supply at the faculty level.
We also complement the empirical strategy based on fixed effects with the instrumental
variables’ approach described above.
3 Data and descriptive statistics
3.1 The History of Italian Universities (HIU) Database
Our main data-source is an original dataset on the History of Italian Universities (HIU).
This is an own compiled register that contains detailed information on institutions providing
higher education in Italy, disaggregated at the faculty level. As mentioned above, this is
the teaching unit, responsible for all BSc or MSc courses within a given disciplinary area.
The dataset covers all years starting from 1861 (year of birth of the Kingdom of Italy) up
to 2010, when the Law 30/12/2010, n. 240 (the so called “Gelmini reform” after the name
of the Public Education Minister in office at that time), which eliminated the faculty from
the governance of public universities.
The primary source we used to construct HIU is Brizzi and Romano (2007), which
report detailed history of Italian universities starting from their foundation onwards. We
integrated this information by several sources on the history of specific universities (see e.g.
Fois, 1991, on the University of Sassari), and faculties (e.g. Silvestri, 2006, on engineering).
We also heavily relied on Gazzetta Ufficiale della Repubblica Italiana, a weekly publication
that collects every public act taken by the government since 1861.15 We double-checked all
information against those provided by open-source archives i.e. Wikipedia, universities’ and
faculties’ websites.16 For the few faculties, for which we found little information on-line, we
contacted the administrative representatives, so as to fill the missing information. Finally,
we validated the data against the actual list of higher education institutions provided by the
Italian Ministry for University Education and Research (MIUR).
The register includes the name of the faculty and its exact address; 15 faculty field
identifiers, which we aggregated into 3 macro-areas of science (Social Sciences, STEM, and
15See Gazzetta Ufficiale della Repubblica Italiana at http : //www.gazzettaufficiale.it/30giorni/concorsi.16Since faculties no longer exist after 2010, their websites are not readily available on the web today. We
retrieved them using Wayback Machine (https : //web.archive.org/), a digital archive of the World WideWeb created by the Internet Archive.
12
Humanities), according to the classification used by the Italian National Statistical Office,
ISTAT. The register includes the year of establishment of the faculty, which is the year when
the faculty is formally established as provider of higher education. HIU also includes name
and address of each university, its governance structure (private or public), and assessments
by the national government (in A,B,C-level). All over-time changes e.g. to the governance
of the faculty or the university, or in the government assessments are also recorded in the
data (See Appendix for details).17
We focus on faculties that deliver standard BSc education i.e. drop from the sample
HE institutions specialised in post-graduate education, and enrolling foreign students only.
Our working HIU sample includes 574 faculties (in 71 universities) registered in the Italian
territory at some point between 1861 and 2010. We use this HIU working sample to construct
a panel cross-province version of the data which follows over-time HE supply in Italian
provinces, during 150 years of Italian history. This is a province by year dataset, which,
for each province i existing at time t, counts the number of faculties present in its territory.
The panel dimension is unbalanced, because the number of provinces changes throughout
the time span. We exclude from the main analysis the first 10 years of Italian history,
as the unification process was still ongoing, featuring the addition of new provinces to the
Italian Kingdom. The final cross-province panel HIU sample includes 11792 observations
for 110 provinces between 1870 and 2010. The main variable is the number of faculties in
province i, as a total and by macro-area of science (Humanities, STEM and Social Sciences).
Following the structure of the register data, we also record information on the total number of
universities, total number of private universities, and total number of A,B,C-level universities
in province i at time t.
Figure 4 gives an historical overview on the total number of faculties in 1870 as compared
to 2010, at the 2010 province level. We observe a major expansion and geographical disper-
sion of higher education, which reflects the pattern of development described in institutional
reports (OECD, 1999; OECD, 2008). As early as 1870, the pre-unitarian distribution fea-
tured a HE supply that was concentrated in the capitals of pre-unitarian states, with the
highest number of faculties in a province being 8 (in Torino, compare Table 1). In 2010, al-
most all provinces have “their own” local HE supply. Those with zero faculties are the most
recent provinces, established as autonomous administrative units in the post-war period. A
few provinces display a HE supply of 15 faculties or more (Bari, Bologna, Milano, Napoli,
17Some of this information is not directly relevant to the present analysis e.g. the dates of some institutionalchanges that occurred at the faculty level. Notice that our data may not map faculty closures as preciselyas faculty openings. However none of our historical sources mentions significant waves of closure of HEinstitutions during Italian history. Also HIU data do not record overtime changes in the addresses ofuniversities and faculties. These information are collected as for 2010.
13
Figure 4: Diffusion of faculties in Italy
Note: authors’ calculation on HIU data
and Roma). All of them include on their territory a metropolitan city.
Figure 4 suggests a spatial component in the expansion of HE supply during Italian
history. We construct several matrices to model spatial interactions in the supply of higher
education between neighbouring provinces. The first one is a contiguity matrix: we define
as neighbours provinces that share the same border, regardless of their distance. A second
matrix is based on linear distances : we define neighbouring provinces on the basis of their
linear distance, regardless of whether they are contiguous or not. For most part of the
analysis, we use and combine these two matrices. We also provide robustness checks using
alternative distance matrices based upon travel distance, and travel time between province
couples (details are in Appendix A.2).
Table 2 presents summary statistics for the main variables. Panel A describes local HE
supply. On average, each province has a supply of over 2.5 faculties, mostly concentrated
in STEM and Social Sciences, and about 0.5 universities, of which the 15% are private, and
over the 75% are classified as A-level.18 In next panels we describe the supply of higher
education of neighbouring provinces. Panel B reports HE supply of “average neighbour”
−i, computed as the average of the HE supply of all neighbours of province i, based on the
contiguity matrix. These figures are obviously very similar to those displayed in Panel A.
A representative neighbour, on average has about 2.9 faculties, mostly in STEM and social
sciences, and roughly 0.5 universities, mostly public and A-level. Panel C reports instead the
HE supply of each individual neighbour j. These figures are very similar to those displayed in
18The high percentage of A-level universities is due to the fact that the De Vecchi-Bottai Reforms, startingfrom 1933 classify all public universities as A-level.
14
Table 2: Summary Statistics
Variable Mean Std. Dev. Min. Max. NPanel A: Local HE Supplyno. of faculties in i 2.515 4.153 0 38 11792no. of humanistic faculties in i 0.479 0.99 0 11 11792no. of stem faculties in i 1.07 1.752 0 9 11792no. of social sciences fac. in i 0.959 1.66 0 19 11792no. of universities in i 0.497 0.842 0 7 11792no. of private universities in i 0.078 0.364 0 4 11792no. of A-level universities in i 0.381 0.621 0 3 11792Panel B: HE Supply of average neighbour, −ino. of faculties in −i 2.884 2.168 0 17.333 11792no. of humanistic faculties −i 0.536 0.544 0 4.333 11792no. of stem faculties in −i 1.24 0.909 0 5 11792no. social sciences fac. in −i 1.098 0.873 0 8.333 11792no. of universities in −i 0.551 0.398 0 2.667 11792no. of private universities in −i 0.089 0.196 0 1.333 11792no. of A-level universities in −i 0.422 0.31 0 1.667 11792Panel C: HE Supply of neighbour, jno. of faculties in j 2.85 4.43 0 38 50803no. of humanistic faculties in j 0.537 1.062 0 11 50803no. of stem faculties in j 1.212 1.837 0 9 50803no. social sciences fac. in j 1.091 1.783 0 19 50803no. of universities in j 0.546 0.895 0 7 50803no. of private universities in j 0.093 0.415 0 4 50803no. of A-level universities in j 0.418 0.639 0 3 50803
Note: authors’ calculation on HIU data
Panels A and B. However, allowing pairwise relationships between provinces i and j, increases
the sample size by roughly four times (as each province i has roughly four neighbours js, on
average), and avoids averaging down the HE supply of neighbour (as shown by comparing
the maximum no. of faculties in Panels B and C).
4 Competition in HE supply
4.1 Empirical strategy
4.1.1 Empirical model
We use our province by year panel dataset to define a model of horizontal spatial interactions
where the supply of higher education F in the i-th Italian province at time t is influenced by
the supply of higher education in the neighbouring provinces. In our baseline model, they
are those provinces that share a border with the i-th province, and may belong to the same
15
region or to a different one.
The spatial interaction can be modelled in several ways, featuring the standard ap-
proaches in the literature. First, we assume that HE supply in province i reacts to HE
supply of its “average neighbour” (Brueckner, 2003). This is the weighted average of HE
supply among the neighbours of the ith province, which is defined as F−it =∑N
j 6=iwijFij,
where wij is a set of weights. We model this relationship through the following linear spatial
competition model:
Fit = α + βF−it + γi + δr(i)t + (Xitφ) + εit, (1)
where β is the coefficient of interest as it captures local strategic interactions among close
provinces. γi and δr(i)t are, respectively, province and region-by-year fixed effects (where
r(i) refers to the region province i belongs to at time t). Xit is a vector of time varying
province level covariates. This includes the total number of universities, number of private
and A-level universities in the province. In several specifications we also include province
level demographic and economic controls that account for potentially spatially correlated
factors.19
Province and region-by-year fixed effects in equation (1) control for a great deal of
unobserved heterogeneity. However, to better account for the omitted determinants of one
province’s HE supply that are spatially correlated with those of the neighbours (e.g. related
to geographical factors), we also estimate a “pairwise” version of equation (1) which takes
the following form (see Parchet, 2014):
Fijt = a+ bFjt + cij + dr(i)t + (Xitf) + eijt. (2)
Equation (2) is estimated for all pairs of neighbouring provinces i-j. Fjt is the HE supply
in each neighbouring province j of province i, cij is the fixed effect of the pair, dr(i)t are the
region-by-year fixed effects and Xit is the usual set of province level controls. The main
advantage of equation (2) relative to (1) is that the inclusion of pair fixed effects allows to
better account for the time-invariant omitted factors that pertain to each i− j couple. Each
couple appears twice, with a given municipality being once on the left-hand side and once
on the right-hand side of the equation. Each pair contributes separately to the estimates,
and b captures the average response of province i to HE supply of neighbour j.
The extensive set of dummies in equations (1) and (2) account for all time-invariant
19In Table 8 below we account for the size of population (total and in the 0-14 age cohort), and size ofthe industry sector. We did not include these variables to the baseline specification as they are availablefrom the Italian Census every ten years, in some cases for the post-war period only. Moreover, due to theirpersistence over time, they are not immune to endogeneity concerns.
16
characteristics at the level of each province, or province pairs. In both specifications, the
region-by-year fixed effects control for all events at the region level that affect all the provinces
in that region simultaneously and identically. The heterogeneity that is left after the inclusion
of the fixed effects is basically the variability over time across provinces within the same
region. Identification of β and b in equations (1) and (2) respectively is obtained through
the comparison of different time-varying patterns in the number of faculties across provinces.
We estimate these two models using various measures of HE supply. Our main indicator
is the total number of faculties in province i and in the neighbourhood (i.e. the average
neighbour −i in equation (1), or each neighbour j in equation (2)). We also distinguish
the extensive margin of HE supply (i.e. having at least one faculty in province i and in
the neighbourhood) from the intensive one (i.e. number of faculties, where HE supply is
available). We alternatively measure HE supply in terms of number of universities available
in province i and its neighbourhood.
The models (1) and (2) are in the standard fixed effects regression format, and can be
easily estimated by OLS. In both models, a key aspect is the definition of neighbouring
provinces, which determines the choice of the matrix of local interactions. As mentioned
above, in the baseline specification, we defined as neighbours of i all provinces that share
a border with the i-th province. In equation (1) this definition of neighbourhood is used
to construct the term F−it =∑N
j 6=iwijFij. Here, weights are settled to reproduce sample
means, i.e. any neighbour is given the same weight: wij = 1N
. Throughout the analysis, we
depart from this baseline matrix and adopt alternative neighbourhood’s definitions based on
e.g. linear distance, travel distance, and travel time measures.
Models (1) and (2) assume contemporaneous spatial competition. This overlooks some
inertia, which may occur in the process of opening a new faculty. As discussed in Section
2 above, several actors are involved in the decision of opening a new faculty, which implies
that the reaction of a province to its neighbours in the supply of higher education is likely
not to be immediate. While it is difficult to define an appropriate lags’ structure, we also
experimented specifications with lagged effects. In our preferred empirical specification, we
allow the spatial competition effect to be modelled as a ten years’ lag: a given province at
time t reacts to the supply of HE in the neighbourhood at time t− 10.
4.1.2 Identification and estimation issues
The main issue in the estimation of equations (1) and (2) arise because the number of faculties
of a neighbouring set of provinces itself depends on the number of faculties of province i:
any change of Fijt may induce each neighbour to adjust Fjt. This gives rise to a reverse
17
causality problem. Moreover, many time-varying determinants of one province’s number
of faculties, such as local economic conditions or local demand for higher education, are
likely to be unobservable and spatially correlated, such that cov(εijt, Fijt) 6= 0. This would
be for example the case of spatially correlated shocks to a neighbouring province. Even if
such unobservable shocks were region specific, they would not be captured by region-by-year
fixed effect unless all neighbours were in the same region as province i. This creates an
omitted variables’ issue that, together with reverse causality may bias the OLS estimation
of (2). The average neighbour model (1) presents the same endogeneity problems, plus
measurement error, due to the fact that the term F−it is an aggregation of the individual
Fijts.
To deal with endogeneity problems we follow an empirical strategy based on instrumental
variables. The key for identification is to isolate variations in number of faculties of competing
provinces that can be plausibly considered exogenous. We propose as instruments the initial
geographical distribution of faculties of each neighbouring province (ICj), interacted with
state-level university reforms (Rt). Our vector of instruments is Zjt = Rt∗ICj, where Rt = 1
if year ≥ t and 0 otherwise. These interactions convey the idea that, while state-level reforms
affect in principle all provinces, the same general reform may have differential effects and
shape HE supply differently, depending on the pre-unitarian endowment of higher education.
As usual, instruments need to satisfy two conditions. First, they have to be exogenous to
HE supply in province i, that is cov(Zjt, εijt) = 0 in equation (2). Second, they should be
relevant, i.e. they should imply enough variation in Fjt. The exogeneity assumption must
hold for both initial conditions and reforms.
The geographical distribution of faculties in pre-unitarian Italian States is most likely
to be exogenous. As discussed in Section 2 above, this reflected the HE policies of pre-
unitarian states which were politically, culturally and linguistically fragmented. They were
often in conflict, and had their own institutions so that they were not coordinating at all
their decisions in any area of public provision, including higher education. As a result of
the extension of the Casati framework to all provinces that progressively joined the Italian
State, the HE system of the Italian Kingdom gradually accredited all the pre-existing HE
institutions, with their specificities without imposing any substantial change. Hence, the
Italian university system was inherited from the past, and the initial distribution of faculties
across provinces was exogenous. While it is true that for any province the initial conditions
are correlated with the subsequent development of the university system (relevance of the
instrument), what is key for identification is that the initial distribution of faculties of the
j and i provinces are uncorrelated. This is indeed the case, for the reasons we discussed in
Section 2 and reprised here.
18
Since initial conditions are fixed over time, they would blur into the pair fixed effects. For
this reason we use law interventions in the university system as another source of exogenous
variation, which is fixed across provinces, but varies over time. We focus on general reforms
of the university system, i.e. laws with a general purpose and not those intended to regulate
some specific need of a limited set of universities. The Gentile reform in 1923 is the first
attempt to create an organic national university system, coherent in its objectives and with
a clear structure. The Devecchi-Bottai reform 1933, together with subsequent ancillary
interventions, created a more overarching and centralised HE system. The 1969 and 1973
reforms liberalised university access to students from technical schools and legislated the
opening of new faculties to cope with the increase in HE demand. The reforms implemented
in the 1980s reorganised the internal governance of universities and the allocation of funding
to higher education. The Bassanini reform in 1997 granted universities more financial and
teaching autonomy.
In this case, the key point for identification is whether (upper-level) state decisions are
arguably exogenous to the (lower-level) province decisions. The exogeneity holds under two
assumptions. The first identifying assumption is that state-level reforms are not driven by
some unobserved time-varying factors that also affect the number of faculties in the province
i and its neighbours. This means assuming that region-by-year dummies capture all the
aggregate component of province-specific shocks. The second requirement is that individ-
ual provinces do not systematically affect state-level higher education policies (no reverse
causality). This rules out the possibility that a province willing to modify its supply of
HE (measured by the number of faculties) has enough policy power to influence the State
decisions about the general setting of the university system. This is likely to be the case
here. On the one hand, provinces are the relevant jurisdictions for decisions on the supply
of local public services such as higher education. On the other hand, they have very limited
legislative competences and their political relevance itself having always been questioned
during Italian history.20 It is rather unlikely that any province has enough “voice” and bar-
gaining power to direct general and national interest reforms that modified the governance
of Italian universities, and implied a general expansion of the university system. On the
Italian territory there is a sufficiently high number of provinces and the population is dis-
persed enough, to prevent specific province from having enough political power to influence
20Petracchi (1962) defines provinces “a big associations of municipalities devoted to the protection of therights of each of them, and to the management of their collective moral and material interests”. However theRepublican Constitutional Law fails to mention their exact tasks and competences. It only states (art. 114)that they are autonomous bodies (as well as municipalities, metropolitan cities, and regions) with their ownstatutes, powers and functions according to the principles established by the Constitution (See also Fabrizzi,2008a,2008b,2008c for historical overviews).
19
state-level decision. Furthermore, we address potential feedback effects from large provinces
to national policy-making, by performing a robustness check where we exclude provinces
hosting a metropolitan city (see Table 8 below).
Interacted with the initial conditions, these reforms are exogenous shocks that modified
the incentives and the net benefits of changing the supply of higher education at the province
level. As we explained above, we associate to each reform a dummy variable which equals
one from the year of the reform onwards and zero before. The idea is that, at any t, each
province and its neighbours face an university system that is the result of the stratification
of current reforms and those inherited from the past. This layering process is measured
by the number of ones in the set of dummies (i.e. the number of laws active) in a given
year. Basically, identification comes by the number of reforms at which at time t a province
has been exposed since the first year it appeared in the sample, interacted by its initial
endowment of faculties.
An interesting feature that emerges from Figure 2, is that the number of faculties in-
creases slowly until 1969, with a sharp and steadily increasing trend emerging afterwards.
On the whole, our 2SLS model exploits one big discontinuity in 1969 plus additional smaller
discontinuities. Post-1969 reforms further increased the autonomy of opening new faculties
at the province level. In the pre-1969 period, the environment was more stable, but, if
anything we expect more effects from the 1923 reform, which increased autonomy and the
opportunity to open new faculties especially in traditional fields like scientific ones.
4.2 Main results
Table 3 presents the estimates of equation (1), based upon the average neighbour approach.
We present four sets of estimates. In Row (a), we estimate the effect of the average number
of faculties in the neighbourhood of province i (i.e. the total number of faculties present in
the neighbouring provinces divided by the number of neighbours) on the total number of
faculties in province i at time t. In Row (b), we isolate the extensive margin of competition
in HE supply i.e. we describe HE supply in province i as a dummy variable, which is equal
to 1 if at least one faculty is present in province i at time t, 0 otherwise. In the same way,
we define the extensive margin of HE supply in the province’s neighbourhood. In Row (c),
we focus on the intensive margin of competition i.e. describe the effect of competition, only
on provinces that have at least one faculty in place in provinces i and j at time t. Finally,
in Row (d) we investigate competition effects when we measure HE supply in terms of the
number of universities (rather than faculties) operating in each province.
20
Table 3: Competition in higher education: average neighbour approach
[1] [2] [3] Obs.(a) average no. of faculties in −i 0.59*** –0.76*** –0.48*** 11383
(0.11) (0.25) (0.14)(b) at least one faculty in −i 0.31*** –0.24* –0.21** 11383
(0.11) (0.12) (0.09)(c) average no. faculties in −i (int. margin) 1.27*** –0.86 –0.47 4064
(0.32) (0.53) (0.30)(d) average no. of universities in −i 0.43*** –0.82*** –0.82*** 11383
(0.10) (0.27) (0.27)province FE yes yes yesregion-by-year FE no yes yesprovincial controls no no yes
Notes:. In Rows (a) and (c) the dependent variable is the total number of majorsin province i. In Row (b) the dependent variable is a dummy equal to 1 if at least onefaculty is active in province i, 0 otherwise. In Row (d), the dependent variable is the totalnumber of universities active in province i. The set of controls include the total numberof universities (not included in specification (d)), the number of A-level universities, andprivate universities in province i. Standard errors clustered by province are reported inparentheses. Significance levels: ∗ : 10% ∗∗ : 5% ∗ ∗ ∗: 1%.
We adopt three different specifications. In Column [1], we present results from simple
OLS, with province fixed effects. In Column [2], we include region-by-year fixed effects. In
Column [3], we add province level controls. Standard errors are always clustered by province.
Results from the preferred specification in Columns [3] show a negative coefficient, generally
significant at conventional levels. Estimates in Row (a) suggest that a one standard deviation
increase in neighbourhood’s HE supply (equivalent to 2.17 faculties in the average neighbour)
is associated with a reduction of HE supply in province i by (2.17 ∗ 0.48 =)1.04 faculty, this
effect being significant at the 1% level. This negative effect is present both on the extensive
and the intensive margin of HE supply in Rows (b) and (c), while becoming less precisely
measured in the latter. The negative effect is also present in Row (d), as we measure HE
supply in terms of no. of universities in each province: an increase in the neighbourhood’s
HE supply by 1 university is associated with a reduction of the local HE supply by over 0.8
universities.
In Table 4 we present estimates of regression model (2) estimated for all pairs of border
provinces (all provinces in the neighbourhood of i). In these estimates pair fixed effects
replaced the province fixed effects. Results confirm that HE supply of neighboring provinces
has a negative effect on the local HE supply. Comparison with average neighbour estimates
in Table 3 suggests that accounting for the pair fixed effects reduces the magnitudes of
the coefficients. Results in Row (a), column (3) suggest that, on average, a one standard
21
Table 4: Competition in higher education: pairwise approach
[1] [2] [3] Obs.(a) no. of faculties in j 0.37*** –0.14** –0.08*** 50803
(0.06) (0.03) (0.04)(b) at least one faculty in j 0.35*** –0.07** –0.04* 50803
(0.06) (0.03) (0.02)(c) no. of faculties in j (int. margin only) 0.73*** –0.16** –0.05*** 22728
(0.15) (0.06) (0.02)(d) no. of universities in j 0.25*** –0.13** –0.13** 50803
(0.06) (0.06) (0.06)provincial pair FE yes yes yesregion-by-year FE no yes yesprovincial controls no no yes
Notes:. In Rows (a) and (c) the dependent variable is the total number of majorsactive in province i. In Row (b) the dependent variable is a dummy equal to 1 ifat least one faculty is active in province i, 0 otherwise. In Row (d), the dependentvariable is the total number of universities active in province i. The set of controlsinclude the total number of universities (not included in specification (d)), the numberof A-level universities, and private universities in province i. Standard errors clusteredby province are reported in parentheses. Significance levels: ∗ : 10% ∗∗ : 5%∗ ∗ ∗: 1%.
deviation increase in HE supply in a neighbouring province (equivalent to 4.43 faculties) is
associated with a reduction of HE supply in province i by (4.43 ∗ 0.08 =)0.35 faculties. This
negative effect is present both on the extensive and the intensive margin of HE supply, and
when we measure HE supply in terms of no. of universities rather than no. of faculties in
each province.
As we discussed above, we resort to 2SLS to tackle endogeneity concerns. Tables 5 and
6 show results for the two stages. In Panels A and B we report estimates for the average
neighbour and pairwise approaches, respectively. We always include the complete set of
fixed effects (FE) and provincial controls, as in column [3] of Tables 3 and 4. We present
different 2SLS FE specifications, as well as their OLS fixed effects (OLS FE) counterpart to
ease comparison. In Column [1], we report estimates from the baseline specification (Row
(a) in Tables 3 and 4). In Column [2], we focus on the 69 “pre-unitarian provinces” that
existed as territorial and administrative units at the moment of the Italian unification. This
is meant to avoid any reverse causality concern e.g. from public policy, to the creation of
new provinces.21 Finally, in Column [3], we acknowledge that the reaction of province i to
HE supply of province j is not instantaneous and allow for a 10 years’ lagged response of
each province to its neighbours.
21In the baseline set of estimates in column [1] we used all 110 provinces. Accordingly, in 2SLS estimateswe set to zero the initial conditions of the 41 provinces that were created during the Italian history.
22
Table 5: 2SLS estimates: initial conditions, reforms and HE supply (1st stage).
[1] [2] [3]Panel A) average neighbour approach(IC−i)*(L. 2102/1923) 0.19 0.21 0.25
(0.16) (0.16) (0.16)(IC−i)*(L. 1592/1933) –0.04 0.24 0.26
(0.10) (0.18) (0.17)(IC−i)*(L. 910/1969) 0.21*** 0.31*** 0.30***
(0.05) (0.08) (0.08)(IC−i)*(L. 766/1973) 0.09** 0.06* 0.06*
(0.04) (0.04) (0.04)(IC−i)*(L. 382/1980) 0.07 0.03 0.03
(0.04) (0.05) (0.05)(IC−i)*(L. 168/1989) 0.04 0.04 0.04
(0.03) (0.05) (0.05)(IC−i)*(L. 245-341/1990) 0.19*** 0.07 0.06
(0.10) (0.10) (0.10)(IC−i)*(L. 59/1997) 0.26*** 0.20 0.09
(0.09) (0.16) (0.12)Observations 11383 8626 8614Panel B) pairwise approach(ICj)*(L. 2102/1923) 0.11*** 0.12*** 0.15***
(0.04) (0.04) (0.04)(ICj)*(L. 1592/1933) 0.10*** 0.13*** 0.15***
(0.03) (0.03) (0.02)(ICj)*(L. 910/1969) 0.22*** 0.25*** 0.24***
(0.02) (0.02) (0.02)(ICj)*(L. 766/1973) 0.05*** 0.05** 0.06***
(0.02) (0.02) (0.02)(ICj)*(L. 382/1980) 0.05*** 0.02 0.01
(0.02) (0.01) (0.02)(ICj)*(L. 168/1989) 0.04*** 0.03*** 0.03***
(0.01) (0.01) (0.01)(ICj)*(L. 245-341/1990) 0.21*** 0.13*** 0.13***
(0.04) (0.03) (0.03)(ICj)*(L. 59/1997) 0.26*** 0.18*** 0.14***
(0.04) (0.05) (0.03)Observations 50803 35510 35060
Notes: First stage of IV FE estimates reported in Table 6.All specifications include the usual set of fixed effects, andprovincial controls. Standard errors clustered by provinceare reported in parentheses. Significance levels: ∗ : 10%∗∗ : 5% ∗ ∗ ∗: 1%.
First stage estimates in Table 5 shows that reforms interacted with the initial HE con-
ditions have a positive impact on HE supply in neighbouring provinces. Significance is
generally higher in Panel B compared to Panel A, due to the smaller standard errors with
pairwise estimation. Summing up the significant coefficients from pairwise estimates suggests
that reform effort that took place during Italian history induced each neighbouring province
23
Table 6: 2SLS estimates: competition in higher education supply (2nd stage).
[1] baseline [2] pre-unitarian [3] pre-unitariansample provinces prov. (lagged 10 y.)
OLS FE 2SLS FE OLS FE 2SLS FE OLS FE 2SLS FEPanel A) average neighbour approachtotal no. faculties in −i –0.48*** –1.14*** –0.09 –1.08** –0.11 –0.94*
(0.14) (0.39) (0.08) (0.49) (0.10) (0.50)Observations 11383 11383 8626 8626 8614 8614K-P rk Wald F-stat 7.167 3.838 3.905K-P rk LM-stat 25.392 15.106 17.199(p-value) 0.001 0.057 0.028Hansen J-stat 7.636 4.017 4.475(p-value) 0.366 0.778 0.724Panel B) pairwise approachtotal no. faculties in j –0.08*** –0.16*** –0.04** –0.13** –0.05* –0.12***
(0.03) (0.05) (0.02) (0.06) (0.02) (0.04)Observations 50803 50803 35510 35510 35060 35060K-P rk Wald F-stat 35.474 25.864 22.231K-P rk LM-stat 61.609 42.057 43.617(p-value) 0.000 0.000 0.000Hansen J-stat 1.644 6.282 3.620(p-value) 0.977 0.507 0.822
Notes:. Regressions in column [1] exploit the baseline sample i.e. 110 provinces during the entireperiod 1871-2010. Regressions in column [2] cover only the set of 69 pre-unitarian provincesfor the entire period. Regressions in column [3] cover pre-unitarian provinces during the entireperiod and take 10 years lags of regressor and controls. Specifications in Panel A include provincefixed effects. Specifications in Panel B include province pair fixed effects. All specificationsinclude region-by-year fixed effects, and the usual set of provincial controls. In 2SLS estimatesthe total no. of faculties in province −i, j is instrumented by the initial conditions (i.e. numberof faculties in −i, j in 1861) interacted by a battery of dummies for higher education reformsin Italy. The full set of coefficients is in Table B-2 below. The full set of first stage coefficientsis in Table 5. Standard errors clustered by province are reported in parentheses. Significancelevels: ∗ : 10% ∗∗ : 5% ∗ ∗ ∗: 1%.
with a pre-unitarian HE supply to open about one faculty. Unsurprisingly, the most relevant
reform is L.910/1969, which liberalised university access.
Table 6 presents results for the second stage. Coefficients estimated under pairwise ap-
proach in Panel B are smaller than their average neighbour counterpart in Panel A. However,
they are also more precisely estimated and more stable across OLS and 2SLS specifications.
F-statistics confirm that instruments have strong predictive power in the first stage of pair-
wise estimates, much less so under the average neighbour approach. The Hansen test con-
firms instruments provide valid exclusion restrictions in the second stage. Pairwise estimates
in column [3] point to sizeable competition effects: HE supply in a province, with e.g. 4
neighbours, decreases by 1.12 faculties, when each neighbour sets up 2 new faculties. This
negative effect can be interpreted as supply of HE services being perceived as close substi-
24
tutes by their consumers (i.e. students) when these are provided in neighbouring provinces.
From now on, for sake of brevity we focus on our preferred specification and report only
pairwise estimates, where all specifications include province pair fixed effects, region by year
fixed effects, and the usual set of controls. 22
4.2.1 Competition across and within field of study
In Table 7, we analyse the issue of competition in HE supply within and between fields of
study. We group all faculties available in our sample into three major groups i.e. Humanities,
STEM (i.e. Science, Technology, Engineering and Mathematics), and Social Sciences.23
We run three sets of regressions for cross and within discipline competition in HE supply.
Table 7 includes three Panels A-C, where our dependent variable is the local supply of HE
in Humanities, STEM, and Social Sciences, respectively. For each panel, we present five
different specifications. In Columns [1] and [3], we perform OLS FE and 2SLS FE estimates,
considering as explanatory variable the neighbour’s HE supply in the own field of study.
In Columns [2] and [4] we include neighbour’s HE supply in the other two fields. Finally,
estimates in Column [5] are identical to those reported in Column [4], but coefficients are
standardised to compare their magnitudes.
Results provide support to the view that HE supply in neighbouring provinces has a
negative effect on the local HE supply within the same field of study. Conversely, there is
not much evidence of negative effects between different fields of study. HE services provided
by neighbouring provinces are perceived as close substitutes within the same field of study,
while it seems difficult to consider e.g. a STEM faculty as a close substitute of a Humanities
faculty. This suggests that differentiating HE supply from that of neighbours can be an
effective policy to avoid competition.
4.2.2 Controlling for HE demand
In Table 8 we refine the analysis and account for factors related to HE demand, which vary
at the level of the province and, if omitted, may affect the estimation of competition effects.
In Panel A, we report the results from our preferred specification (Table 6, Panel B, Column
[3]) to ease comparability of results. In Panel B, we check whether our results are driven
22Estimates with the average neighbour approach are qualitatively similar, but, for the reasons explainedabove, in general less precise. They are available upon request by the authors.
23Humanities include Education, Linguistic Studies, Literature, and Psychology. The STEM group in-cludes Agricultural Studies, Architecture, Chemistry and Pharmacy, Geology and Biology, Engineering andScientific Studies. Social Sciences include Medical Studies, Economics and Statistics, Law, and Socio-PoliticalStudies.
25
Table 7: Cross-disciplinary competition: humanities, stem, social sciences
[1] [2] [3] [4] [5] [6]OLS FE OLS FE 2SLS FE 2SLS FE 2SLS FE Obs.
Panel A) Humanities 32644facultiesj , Humanities –0.06* –0.07* –0.02 –0.07* –0.08*
(0.04) (0.04) (0.07) (0.04) (0.04)facultiesj , STEM 0.01 –0.02 –0.03
(0.02) (0.03) (0.05)facultiesj , Social Sciences 0.01 –0.00 –0.01
(0.02) (0.01) (0.02)K-P rk Wald F-stat 4.879 11.141 11.141K-P rk LM-stat 20.558 41.826 41.826(p-value) 0.008 0.007 0.007Hansen J-stat 7.641 24.824 24.824(p-value) 0.365 0.255 0.255Panel B) STEM 32644facultiesj , STEM –0.09*** –0.11*** –0.08* –0.05** –0.09**
(0.03) (0.03) (0.04) (0.02) (0.04)facultiesj , Humanities 0.02 –0.03 –0.03
(0.02) (0.02) (0.02)facultiesj , Social Sciences 0.02 0.00 0.00
(0.02) (0.01) (0.02)K-P rk Wald F-stat 24.294 11.141 11.141K-P rk LM-stat 33.890 41.826 41.826(p-value) 0.000 0.007 0.007Hansen J-stat 4.878 15.791 15.791(p-value) 0.675 0.781 0.781Panel C) Social Sciences 32644facultiesj , Social sciences –0.04* –0.04 –0.16** –0.07** –0.12**
(0.02) (0.03) (0.07) (0.03) (0.05)facultiesj , STEM 0.02 –0.01 –0.03
(0.03) (0.02) (0.04)facultiesj , Humanities –0.01 0.03 0.03
(0.04) (0.02) (0.03)K-P rk Wald F-stat 23.301 11.141 11.141K-P rk LM-stat 34.300 41.826 41.826(p-value) 0.000 0.007 0.007Hansen J-stat 2.790 20.361 20.361(p-value) 0.904 0.498 0.498
Notes:. Baseline pairwise estimates as in Table 6, Panel B, Column [3]. OLS FE estimates inColumns [1] and [2]. 2SLS FE estimates in Column [3] with one endogenous regressor i.e. theno. of faculties in the respective discipline (Humanities in Panel A, STEM in Panel B, SocialSciences in Panel C), in the neighbouring province. In Column [4] 2SLS FE estimates withthree endogenous regressors i.e. the no. of faculties in Humanities, STEM and Social Sciencesin the neighbouring province. The specification in Column [5] is the same as in Column [4], butregressors are standardised to have zero mean and unity standard deviation. In 2SLS estimates,the instruments are interactions of initial conditions (total no. of faculties in each discipline in1861) with a battery of dummies for higher education reforms in Italy. See Table B-3 below forthe first stage estimates. All specifications include provincial pair fixed effects, region-by-yearfixed effects, the usual set of provincial controls for the local province. Standard errors clusteredby province are reported in parentheses. Significance levels: ∗ : 10% ∗∗ : 5% ∗ ∗ ∗:1%.
26
Table 8: HE supply Competition: local demand effects
[1] [2] [3]OLS FE 2SLS FE obs.
Panel A) baseline specification 34912total no. faculties in j –0.05* –0.12***
(0.02) (0.03)K-P rk Wald F-stat 22.241K-P rk LM-stat 43.612p-value 0.000Hansen J-stat 3.635p-value 0.821Panel B) drop metropolitan cities 22814total no. faculties in j –0.11** –0.22***
(0.04) (0.07)K-P rk Wald F-stat 37.514K-P rk LM-stat 25.099p-value 0.001Hansen J-stat 7.139p-value 0.415Panel C) control for population size 34912total no. faculties in j –0.04 –0.12***
(0.03) (0.04)K-P rk Wald F-stat 20.689K-P rk LM-stat 40.629p-value 0.000Hansen J-stat 3.328p-value 0.853Panel D) control for participation and size of the industry sector 34912total no. faculties in j –0.05* –0.12***
(0.02) (0.04)K-P rk Wald F-stat 22.759K-P rk LM-stat 43.717p-value 0.000Hansen J-stat 3.779p-value 0.805Panel E) control for share of 0-14 years old 11442total no. faculties in j –0.05* –0.09*
(0.03) (0.05)K-P rk Wald F-stat 31.930K-P rk LM-stat 40.092p-value 0.000Hansen J-stat 2.576p-value 0.765Panel F) placebo: “alphabetical” neighbors 48337total no. faculties in j 0.009 0.063
(0.020) (0.061)K-P rk Wald F-stat 76.556K-P rk LM-stat 44.544p-value 0.000Hansen J-stat 8.124p-value 0.322
Notes: In Panel A baseline pairwise estimates as in Table 6, Panel B, column [3]. InPanel B, provinces with metropolitan cities (Bari, Bologna, Cagliari, Catania, Firenze,Genova, Messina, Milano, Napoli, Palermo, Reggio Calabria, Roma, Torino, Verona)are dropped from the sample. In Panel C, the total population is included in the setof controls. In Panel D, the share of active people in the industry sector, and theparticipation rate are included among the controls. In Panel E, the share of populationbelonging to the 0-14 cohort is included among the controls (available only for period1951-2010). In Panel F, neighbors are defined as provinces whose name starts withthe same letter of the alphabet. All specifications include provincial pair fixed effects,region-by-year fixed effects, and the usual set of provincial controls. In 2SLS estimates,the instruments are interactions of initial conditions with higher education reforms.The full set of coefficients is in Table B-4 below. Standard errors clustered at theprovince level. Significance levels: ∗ : 10% ∗∗ : 5% ∗ ∗ ∗: 1%.
27
by the 14 Italian metropolitan cities.24 Some of these cities (e.g. Milano) were not primary
university centres in the 19th century, but became so only afterwards as a result of their
population growth. We re-run our regressions after excluding them and our results are not
altered. In the same spirit, in Panel C, we check how our results change if we include as a
control the total population (in log).25 Our results survive to its inclusion.
Demand for higher education may actually follow the gradual expansion of the industrial
sector, and labour market participation during the 20th century. To account for that, in
Panel C we include as controls the participation rate and the population share active in
the industry sector. Also the inclusion of these controls makes no difference. Demographic
changes in the composition of population may also determine changes in HE demand. In
Panel D, we include the population share in the 0−14 cohort. We were able to recollect this
information only from 1951 onwards, and with some missing information for some provinces.
For this reason the inclusion of this control produces a substantial loss of information. Our
results however survive also to this robustness check. Finally, one may argue that our
results are driven by some underlying omitted factors, which we are not able to control.
If this was the case, the estimated competition effect would be likely to survive, whatever
the neighbourhood relationship hypothesised. In Panel F we perform a placebo exercise
imposing a non-sense alphabetical neighbourhood relationship: provinces are considered
neighbours if their name starts with the same letter of the alphabet, regardless of their
actual neighbourhood relationships (e.g. Alessandria is coded as neighbour of Agrigento,
despite being over 1000 Kms aparts, and other provinces starting with an “A” only). The
estimates show that the negative coefficient of the number of faculties in j disappears. This
reassures us that the competition effect is specific to the neighbourhood matrix we have
chosen.
4.2.3 The spatial reach of competition effects
Up to now we defined neighbours the provinces who share a border. Still, even provinces that
do not share a border can be very close to each other in terms of distance. We used available
Google Maps applications to compute alternative matrices based upon linear distance, travel
distance and travel time between province capitals (see Appendix A.2 for details). Estimates
using these alternative matrices allow us to discuss the “spatial reach” of competition, that
24These are identified by the Italian Constitution: Bari, Bologna, Catania, Firenze, Genova, Messina,Milano, Napoli, Palermo, Reggio Calabria, Roma, Torino, Verona.
25Notice that population data are drawn from the Italian Census data, which occurred about once everyten years during the period 1861-2011. We assigned each population-by-province data point to the successiveyears, until a new wave of census data is available.
28
Tab
le9:
Com
pet
itio
nin
HE
supply
:linea
rdis
tance
mat
rix
(in
Km
)an
dsp
atia
lre
ach
[1]
wit
hin
90K
m[2
]b
etw
een
90
an
d180
Km
[3]
bet
wee
n180
an
d270
Km
[5]
bet
wee
n270
an
d360
Km
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
tota
ln
o.fa
cult
ies
inj
–0.0
39**
–0.0
95**
–0.0
45***
0.0
03
0.0
03
–0.0
14
–0.0
13
0.0
01
(0.0
18)
(0.0
39)
(0.0
14)
(0.0
17)
(0.0
07)
(0.0
10)
(0.0
11)
(0.0
10)
Ob
serv
atio
ns
4350
043
500
88512
88512
99184
99184
77114
77114
K-P
rkW
ald
F-s
tat
45.8
50
149.6
15
167.0
54
85.4
79
K-P
rkL
M-s
tat
47.3
64
49.4
65
51.6
81
53.4
63
p-v
alu
e0.
000
0.0
00
0.0
00
0.0
00
Han
sen
J-s
tat
9.19
04.3
49
7.1
42
10.2
33
p-v
alu
e0.
239
0.7
39
0.4
14
0.1
76
Note
s:.
Bas
elin
ep
airw
ise
esti
mat
es.
In2S
LS
esti
mate
s,th
eto
tal
no.
of
facu
ltie
sin
pro
vin
cej
isin
stru
men
ted
by
the
init
ial
con
dit
ion
sin
tera
cted
by
hig
her
edu
cati
onre
form
sin
Italy
.S
eeT
ab
leB
-7b
elow
for
det
ail
s.S
tan
dard
erro
rscl
ust
ered
by
pro
vin
ceare
rep
ort
edin
par
enth
eses
.S
ign
ifica
nce
leve
ls:
∗:
10%
∗∗:
5%
∗∗∗:
1%
.
29
is the distance up to which HE supply in a province responds to HE supply set elsewhere.
In Table 9 we report estimates using the definition of neighbourhood based on exogenous
linear distance between provinces. Columns [1] to [5] show how HE supply of each province
reacts to the HE supply of provinces located at different linear distances, on average. Results
suggest that HE supply in a province reacts more to the supply of closer provinces. The effect
is concentrated within a linear distance of 90 kilometres. Negative coefficients appear also for
higher distances, however the effects are much smaller in size, and generally non statistically
significant. Any evidence of HE supply competition disappears for linear distances beyond
270 kilometers.
In Table B-5 in the appendix, we report results for matrices based on travel distance
and travel time. In Panels A-B of Table B-5 we define neighbours on the basis of the travel
distances, and travel time between the provinces’ capitals. Using these alternative measures
we find that competition effect is concentrated among a travel distance of 120 kilometres and
a travel time of 80 minutes. The spatial reach of strategic interactions can thus be estimated
to be somewhat within these ranges. The effect disappears for travel distances beyond 360
kilometers and travel time exceeding 3 hours.
4.2.4 Heterogeneous effects
We perform three additional exercises to investigate the heterogeneity in the competition of
HE supply. First, we report the results dividing the sample into two periods across World
War II (WWII): a pre-war period is defined as before year 1936, while post-war as after year
1950. Results, reported in Panel A of Table 10, confirm that HE supply in neighbouring
provinces has a negative coefficient both for pre-war period (Column [1]) and the post-war
period (Column [2]), although being less precise in the pre-war period.
Second, to capture long-lasting differences in development of HE system within Italy we
also perform an analysis by two macro regions. In Panel B, we present results for Northern
regions which belong to NUTS1 North-East and North-West (Column 1), and Centre-South
regions belonging to NUTS1 Centre, South and Islands (Column 2). The estimates show
that the negative effect is still present in both Columns [1] and [2], being more precise for
2SLS FE compared to OLS FE estimates.
Finally, estimates in Table 6 consider homogeneous interactions between neighbouring
provinces, regardless of the NUTS2 region they belong to. In Panel C we analyse hetero-
geneous interactions between provinces that belong to the same region (Column [1]), and
provinces that belong to different regions (Column [2]). Interestingly enough, the competi-
30
Table 10: HE supply Competition: heterogeneity
Panel A) pre-WWII vs. post-WWII period[1] pre-war [2] post-war
OLS FE 2SLS FE OLS FE 2SLS FEtotal no. faculties in j –0.03 –0.11 –0.04* –0.10**
(0.02) (0.09) (0.02) (0.05)Observations 17283 17283 14330 14330K-P rk Wald F-stat 32.639 30.752K-P rk LM-stat 19.372 41.734p-value 0.000 0.000Hansen J-stat – 2.637p-value – 0.756Panel B) North vs. Centre-South
[1] North [2] Centre-SouthOLS FE 2SLS FE OLS FE 2SLS FE
total no. faculties in j –0.05 –0.14** –0.04 –0.09**(0.03) (0.06) (0.03) (0.04)
Observations 17785 17785 17275 17275K-P rk Wald F-stat 19.167 39.900K-P rk LM-stat 22.602 22.540p-value 0.004 0.004Hansen J-stat 4.948 6.881p-value 0.666 0.441Panel C) intra-regional vs. inter-regional competition
[1] intra-regional [2] inter-regionalOLS FE 2SLS FE OLS FE 2SLS FE
total no. faculties in j –0.11** –0.18*** 0.03 –0.06(0.04) (0.06) (0.04) (0.07)
Observations 20234 20234 14687 14687K-P rk Wald F-stat 21.651 12.205K-P rk LM-stat 37.903 24.197p-value 0.000 0.002Hansen J-stat 5.331 1.947p-value 0.620 0.963
Notes: Baseline pairwise estimates. In Panel A, pre-WWII andpost-WWII periods are defined as before 1936 and after 1950, respec-tively. In Panel B, North regions belong to NUTS1 “North-East and“North-West”, while Centre-South regions belong to NUTS1 “Centre”,“South” and “Islands”. In Panel C, intra-regional competition is be-tween provinces in the same NUTS2 region, while inter-regional com-petition is between provinces belonging to different NUTS2 regions. In2SLS estimates, the instruments are interactions of initial conditions(total no. of faculties in 1861) with higher education reforms in Italy ineach sub-period. See Table B-6 below for the first stage estimates. Allspecifications include provincial pair fixed effects, region-by-year fixedeffects, and the usual set of provincial controls. Standard errors clus-tered at the province level. Significance levels: ∗ : 10% ∗∗ : 5%∗ ∗ ∗: 1%
31
tion effect is mostly concentrated within the same region. This suggests that substitutability
between HE institutions is lower when these are located in different regions, even though
they are in neighbouring provinces.
4.3 Sensitivity analysis
4.3.1 Alternative definitions of pre-unitarian provinces
In our preferred specification, we define as pre-unitarian provinces that existed as adminis-
trative units at the onset of Italian history. However, we allow their territory and borders
to change overtime, alongside the creation of new provinces during Italian history.
We take this modelling choices to be as precise as possible in the measurement of the
local neighbourhood effects. However this is itself not immune to criticisms. It is well
known that early Italian governments re-aggregated most provinces of pre-unitarian states
in larger ones on the basis of political, economic, administrative and demographic consid-
erations (Palombelli, 2012). Thus, our definition of pre-unitarian province may not fully
serve the purpose of exogeneity, provided that aggregation might be carried out considering
also the territorial provision of HE services. It may also be argued that it is not only the
administrative dimension of province that matters, but the territorial subdivision. In that
case exogeneity would require to restore provinces’ pre-unitarian territories and borders.
In Table 11 below, column [1] we adopt a definition of pre-unitarian province that encom-
passes not only their existence as administrative units, but also their territory and borders in
1870. In order to do so, we assign all faculties set up in a province created during Italian his-
tory to the territory of the corresponding pre-unitarian province. We define neighbourhood
relationship accordingly, on the basis of these pre-unitarian borders, which we maintain con-
stant during Italian history.26 Our main results are preserved. However, the decrease in size
and significance of OLS estimates suggests the adoption of this definition of pre-unitarian
province may induce some measurement error. In the next two columns of Table 11 we build
upon the consideration that the 41 provinces created during Italian history are the same
provinces of pre-unitarian states that were downgraded to districts of the 69 we use in the
baseline analysis (Palombelli, 2012). If this is the case, the process of provinces’ creation
26To make some examples, the faculties of the province of Pescara belong to the pre-unitarian province ofChieti (as the corresponding territory was then assigned to the new-born province of Pescara in 1927), whichwe assume always shares borders with the pre-unitarian territories of l’Aquila, Teramo and Campobassoduring Italian history. Similarly we assign HE supply set up in the province of Taranto to the pre-unitarianterritory of Lecce (to which the province of Taranto belongs), in the neighbourhood of the pre-unitarianprovinces of Bari and Potenza (which shared borders with Lecce in 1870, but never with Taranto).
32
Table 11: Competition in HE supply: alternative definitions of pre-unitarian provinces
[1] existing in 1870, [2] existing in 1870, [3] existing in 2010borders at 1870 borders in 2010 borders in 2010
OLS FE 2SLS FE OLS FE 2SLS FE OLS FE 2SLS FEtotal no. faculties in j –0.01 –0.09** –0.02 –0.07* –0.08*** –0.09***
(0.02) (0.04) (0.02) (0.04) (0.02) (0.03)Observations 42813 42813 42813 42813 68410 68410K-P rk Wald F-stat 36.130 41.221 48.375K-P rk LM-stat 42.550 44.206 71.220p-value 0.000 0.000 0.000Hansen J-stat 6.046 6.797 10.327p-value 0.534 0.450 0.171
Notes:. Baseline pairwise estimates. Borders and territory of pre-unitarian provinces aremaintained constant over the entire period 1870− 2011. In column [1] pre-unitarian provincesare defined by borders and territory at 1870 i.e. the HE supply of the territory of provincescreated during Italian history is reassigned to the pre-unitarian province from which the newprovinces have been created. In column [2] pre-unitarian provinces are defined by borders andterritory in 2010. In column [3] pre-unitarian provinces are defined as administrative units in2010 at their 2010 borders and territories. Standard errors clustered by province are reportedin parentheses. Significance levels: ∗ : 10% ∗∗ : 5% ∗ ∗ ∗: 1%.
during Italian history actually features these 41 districts gaining autonomy again i.e. restores
the true pre-unitarian distribution. In column [2] we define pre-unitarian provinces the ad-
ministrative units that existed in 1870, but at their 2010 borders. In column [3], we define
pre-unitarian provinces all administrative units that existed in 2010 at their 2010 borders.
In both cases, we maintain these definitions constant over the 150 years of Italian history.
Our results are again confirmed alongside size and significance of the estimated coefficients.
4.3.2 Alternative empirical strategies
In Table 12, we check the sensitivity of our results to the adoption of alternative empirical
strategies. In column [1], we perform a robustness check in the spirit of De Giorgi et al.
(2010) and include in the set of instruments the 2nd degree spatial lag of the HE supply of
the local province. The identifying assumption is that all provinces z that share a border
with province j, but not with province i, are valid instruments to identify the impact of
HE supply of province j on local supply in i. Our main results are preserved however the
Hansen J-test rejects the exogeneity of the instruments. This suggests that there are other
factors over and beyond sharing a border that matter in determining local interactions.
These factors undermine the validity of this instrument in the present setting. In column
[2], we concentrate on the decade 1966-1976 only. In this way, we focus on an “event
study” that uses the major liberalisation entailed by the 1969 and 1973 reforms to identify
33
Table 12: Sensitivity analysis: alternative empirical strategies
[1] 2SLS FE [2] 2SLS FE [3] 2SLS FE [4] 2SLS FE [5] 2SLS FE [6] 2SLS FEtotal no. faculties in j –0.12** –0.19* –0.14*** –0.13*** –0.16*** –0.12**
(0.06) (0.11) (0.05) (0.05) (0.05) (0.05)Observations 109817 2653 34904 35060 32210 35328K-P rk Wald F-stat 29.118 27.529 28.608 58.178 23.230 63.414K-P rk LM-stat 39.760 23.703 42.723 44.423 41.329 58.088p-vale KP 0.000 0.000 0.000 0.000 0.000 0.000Hansen J-stat 17.212 0.014 2.540 3.823 4.908 7.434p-value Hansen 0.028 0.906 0.924 0.800 0.427 0.385
Notes: Baseline pairwise estimates. In column [1], the set of instruments includes the 2nd degree spatial lag ofthe local province. Regressions in column [2] cover the sub-period 1966-1976 only. In column [3] region-by-yearFE for the region of the neighbouring province are included. In column [4], we include the number of universities(total, A-level, private) of the neighbouring province as additional controls. In column [5], we lag instruments tenyears relative to the dependent variable in the first stage. In column [6] we define neighbours as provinces thathave a shared border, within the spatial reach of 90 Km. Standard errors clustered by province are reported inparentheses. Significance levels: ∗ : 10% ∗∗ : 5% ∗ ∗ ∗: 1%.
competition effects.27 The drop in the number of observations reduces the precision of the
estimates, however our results are not altered. In columns [3] and [4], we include region-by-
year dummies of the neighbouring province, and its HE supply controls, respectively. Also
in this case our results are confirmed. Results do not change in column [5] either, as we
consider a ten years lagged effect of the instrument on the HE supply of the neighbour in
the first stage. Finally, in column [6] we consider competition between provinces that share
a border and within a spatial reach of 90 Km. Our results are confirmed in this case too.
5 Economic value of HE supply
In this section, we analyse the economic value of HE supply. Besides being an interesting
exercise per se, this is also useful to compute the opportunity cost of competition forces,
in terms of value added per capita that is lost when not setting up a local faculty due to
competition forces.
To this purpose, we estimate the following model:
ln(Yit) = α + βFit−10 +X ′itγ + δi + µr(i),t + εit. (3)
27This exercise can also be viewed as an additional robustness check that our results are not determinedby omitted time varying factors associated with overtime changes to territory and borders of provinces. Infact the number and borders of provinces remain constant during this decade, the only exception being thecreation of the Province of Isernia in 1971.
34
where Yit is (log) value added per capita in province i, at time t; Fit is the total number
of faculties in province i at time t; Xit is a vector of province level controls, i.e. the rate
of population growth, the participation rate and the size of the industry sector in province
i at time t (as measured by the share of active population in the industry sector).28 δi is a
province fixed effect, and µr(i),t is a region-by-year FE. Finally, εit is the error term.
Our main parameter of interest is β: this measures the marginal value of one faculty in
terms of local value added per capita. Two endogeneity issues complicate the estimation of
equation (3). First, reverse causality going from economic performance to HE supply: richer
and/or more productive provinces may express a larger (or smaller) demand of HE. Thus,
the opening of a new faculty may result from this demand. Second, omitted factors may
motivate both an increase in value added per capita and the opening of a new faculty.
We try and address these concerns implementing the same instrumental variables esti-
mator, which uses pre-unitarian supply of higher education in the province, interacted with
state-level reforms Zit = ICi∗R′t as an instrument for Fit (both lagged ten years) in equation
(3). We already discussed extensively the exogeneity of initial conditions and the reforms.
Being prior to the formation of the country, and specific to the several pre-unitarian States
present on the Italian territory, the pre-unitarian HE supply is pre-determined, thus not
affected by the demand for education that emerged after the unification (Squicciarini and
Voigtlander, 2015). The additional identifying assumption we need for Zi being a valid
instrument for Fit in equation (3) is that, conditional on the large set of province and region-
by year fixed effects, the interactions between state level reforms and provinces’ HE initial
conditions, do not have a direct effect on the local value added per capita.
Information on value added, population, and participation come from the Italian Census
Data, which are collected roughly every 10 years. This means that we exploit a ten years’
variation to estimate equation (3). During the 150 years’ sample period, we observe provinces
at most 16 times, i.e. T = 16 in our unbalanced panel. This changes the interpretation of the
state-level reforms. In some cases R′t captures exposure to “reform packages” implemented
during the decade t i.e. R′t = 1 if decade > t, 0 otherwise.29
Table 13 reports results from various specifications of model (3). We start by presenting
the effect of local HE supply on economic performance, i.e. we consider the HE supply of
28In their analysis of the HE determinants of the industrial revolution in Prussia, Becker et al. (2011)consider the share of population active in the industry sector as the main outcome variable. We use it ratheras a control, as our time period is posterior to the industrial revolution.
29More precisely the dummy R21/30 includes only the L. 2102/1923. Similarly R31/40 includes the L.1592/1933 and R61/70 includes the Law 910/1969. Conversely R71/80 includes a package composed by bothLaws 766/1973 and 382/1980. Similarly, R81/90 includes L. 168/1989 and Laws 245-341/1990. Finally R91/01
covers Laws 59-127/1997.
35
Tab
le13
:E
conom
icva
lue
ofH
Esu
pply
and
exte
rnal
itie
s
[1]
OL
S[2
]2S
LS
FE
[3]
2S
LS
FE
[4]
2S
LS
FE
[5]
2S
LS
FE
[6]
2S
LS
FE
[7]
2S
LS
FE
[8]
2S
LS
FE
[9]
2SL
SF
EL
IML
LIM
LL
IML
Fac
ult
ies
ini
0.00
77**
*0.
0070
*0.0
081*
0.0
133***
0.0
108***
0.0
144**
0.0
109**
0.0
114***
0.0
112***
(0.0
017)
(0.0
041)
(0.0
042)
(0.0
046)
(0.0
040)
(0.0
055)
(0.0
044)
(0.0
022)
(0.0
021)
Fac
ult
ies
in−i
0.0
073**
0.0
063**
0.0
087*
0.0
069**
0.0
029**
0.0
036***
(0.0
035)
(0.0
028)
(0.0
045)
(0.0
032)
(0.0
012)
(0.0
013)
Ob
serv
atio
ns
937
937
937
937
937
937
937
874
874
−i
wit
hin
90K
m–
––
no
yes
no
yes
no
yes
Inst
rum
ents
no
IC*R
IC*R
IC*R
IC*R
IC*R
IC*R
IC*R
IC*R
,Ft−
20
Ref
orm
s(R
)–
all
thre
eth
ree
thre
eth
ree
thre
eth
ree
thre
eK
-Prk
Wal
dF
-sta
t–
12.6
5911.8
01
3.0
11
8.8
72
3.0
11
8.8
72
49.4
39
69.8
57
K-P
rkL
M-s
tat
–25
.160
23.0
65
17.4
37
15.2
19
17.4
37
15.2
19
24.8
48
24.0
58
p-v
alu
eK
P–
0.00
00.0
00
0.0
04
0.0
09
0.0
04
0.0
09
0.0
01
0.0
01
Han
sen
J-s
tat
–13
.742
4.6
01
5.2
86
5.8
90
5.1
42
5.8
95
7.7
10
6.8
21
p-v
alu
eH
anse
n–
0.01
70.1
00
0.2
59
0.2
08
0.2
73
0.2
07
0.2
60
0.3
38
Note
s:In
all
spec
ifica
tion
sth
ed
epen
den
tva
riab
leis
valu
ead
ded
per
cap
ita
inlo
gte
rms.
Th
eto
tal
nu
mb
erof
facu
ltie
sin
the
pro
vin
cean
din
the
nei
ghb
orh
ood
are
lagg
ed10
year
s.A
llsp
ecifi
cati
on
sin
clu
de
as
contr
ols
,gro
wth
inlo
cal
pop
ula
tion
,th
ep
art
icip
ati
on
rate
an
dth
esi
zeof
the
ind
ust
ryse
ctor
.A
llsp
ecifi
cati
ons
incl
ud
ep
rovin
ceF
Ean
dre
gio
nby
year
du
mm
ies.
Insp
ecifi
cati
on
s[3
]-[9
]th
eth
ree
refo
rmp
ack
ages
inve
ctorR
are
R21/30,R
31/36,
andR
21/30
only
.S
tan
dar
der
rors
clu
ster
edat
the
pro
vin
cele
vel
inp
are
nth
eses
.S
ign
ifica
nce
leve
ls:
∗:
10%
∗∗:
5%
∗∗∗:
1%
.
36
the local province the only determinant of local economic prosperity. In columns [1] and [2]
we report OLS and 2SLS results. These results point to positive returns from HE supply,
of very similar magnitudes in OLS and 2SLS estimates. A concern with these estimates is a
likely collinearity between our instruments and the country-by-region dummies, due to the
ten years’ span of the time dimension available for this set of estimates. First stage estimates
in Table B-9 confirm this suspect, as in our vector of reform packages only the first three
have actual explanatory power in the first stage. From columns [3] to [9], we accordingly
report 2SLS FE estimates, using only the relevant instruments in the first stage estimate of
column [2] i.e. ICi ∗R21/30, ICi ∗R31/36, and ICi ∗R61/70. This greatly mitigates the problem
of collinearity.
Estimates in columns [1] to [3] do not take into account the fact that opening new
faculties in the neighbourhood may actually produce direct externalities on local economic
activity. In columns [4] to [9] we investigate this possibility, by including either an externality
effect coming from all the neighbourhood (columns [4], [6], and [8]) or an externality effect
coming from the neighbours within the spatial reach of 90 Km (columns [5], [6], [7]). In
these estimates, the F-test statistic of the additional instruments is below the critical value
of 10. Accordingly, in columns [5], [7], and [9] we estimate the model by limited information
maximum likelihood (LIML), which is median-unbiased in over-identified models. Also,
in column [9] we include as instruments the 20 years lagged values of HE supply in the
local province and neighbourhood, which increases the power of instruments in the first
stage. Estimated coefficients in columns [4]-[9] are larger in size, more robust and precisely
estimated than in columns [1]-[3].
Overall, results in Table 13 confirm the positive local returns of HE supply. Taken at their
face value, these estimates suggest that, on average, one faculty more in the local province
raises value added there by more that 1%. We can interpret this value as the local opportunity
cost of not opening one faculty in the province. As we discussed in Section 4, this is roughly
the result of competition coming from e.g. 4 neighbouring provinces, each one opening 2
faculties. However, our estimates also suggest that 8 faculties in the neighbourhood produce
direct positive externalities on local value added per capita by almost (8 ∗ 0.3) = 2.4%. This
implies that direct positive externalities coming from faculties located in the neighbourhood,
more than compensate the local costs of competition effects.
Two remarks are in order. First of all, our results are in line with those from existing
studies that use a similar empirical approach (Valero and Van Reenen, 2016). This is non
negligible, as we look at one country only (Italy), a more disaggregated regional unit (NUTS3
instead of NUTS2), and a different measure of HE supply (faculties instead of universities).
37
Second, our results do not allow pointing out the exact channel that determines the effect
of HE supply on value added per capita. Our results can be determined by either human
capital accumulation (Ciccone and Peri, 2006; Bratti and Leombruni, 2014) or long-term
changes in the industry composition and technologies (see e.g. Ciccone and Peri, 2011), or
both. Pointing out the exact channel is left as an interesting subject for further research.
6 Conclusions
In this paper, we use an own-built historical dataset on the formation of higher education
institutions in Italy (History of Italian Universities - HIU) to investigate local competition
in the supply of higher education.
To this purpose, we use HIU data to analyse whether the existence of a higher education
institution in a neighbouring province hinder the local supply of higher education. Because
of spatial correlation and reverse causality issues, we estimate local competition effects, by
implementing an IV strategy which exploits exogenous variation associated with the initial
conditions of the Italian higher education system interacted with the most comprehensive
reforms of higher education that took place over Italian history.
We find evidence that the expansion of HE institutions during Italian history, particu-
larly post-WWII (OECD, 1999, 2008) was deeply shaped by higher education reforms and
pre-unitarian supply. The main set of estimates reveals non-negligible local competition
effects: on average a province which has 8 faculties in the neighbourhood reduces its local
supply by more than 1 faculty. We also perform separate analyses and found that such
local substitutability between faculties is mostly concentrated within the same field of study
(humanities, stem, social sciences), within the same NUTS2 region, and in a spatial reach of
90 Km.
These findings have important implications for the higher education policy of developed
countries. They suggest that local substitutability of higher education supply may act as
a remarkable “discipline effect” to HE expansion from below. Starting from the 1960s, re-
gional HE supply considerably increased in OECD countries (OECD, 1999). Based on our
estimates, the cost associated with the ensuing HE competition may be non negligible in
economic terms; however, it tends to be overcompensated by direct positive externalities
coming from the neighbourhood. Overall, our findings suggest that national OECD gov-
ernments should be willing to regulate, to favour an efficient provision of higher education
services. The expansion of local higher education providers should be restrained, especially
in smaller provinces, surrounded by metropolitan areas, which are subject to the strongest
38
competition effects. Differentiating HE supply (e.g. by field of study) can be a useful device
to mitigate competition intensity, and feature an efficient design of local provision of higher
education.
Acknowledgements
We would like to thank the Department of Economics and Finance, Universita Cattolica del
Sacro Cuore (UCSC) for selecting this project for the Research Assistance Program. We are
particularly grateful to Ginevra Gallassi, Roberta Terranova, Giuseppe Grasso, and George
Jacob for precious assistance at various stages of our research. We would also like to thank
the personnel from the library “G. Billanovich, UCSC” for their valuable help in finding
out materials used to build the HIU dataset and the Istituto Tagliacarne for the data on
province value added. We finally wish to thank colleagues who attended presentations at the
internal seminar of UCSC; the 2017 LEER Workshop in Leuven; 2017 IWAAE Workshop
in Catanzaro; 2017 APET Conference in Paris; 2017 AIEL Conference in Cosenza, and the
invited RUE seminar at the School of Economics of the University of Barcelona for very
helpful discussions. The usual disclaimer applies.
39
References
Aghion, P., Dewatripont, M., Hoxby, C., Mas-Colell, A., and Sapir, A. (2010). The gov-
ernance and performance of universities: evidence from europe and the us. Economic
Policy, 25(1):7–59.
Becker, S. O., Hornung, E., and Woessmann, L. (2011). Education and catch-up in the
industrial revolution. American Economic Journal: Macroeconomics, 3(3):92–126.
Braga, M., Checchi, D., and Meschi, E. (2013). Educational policies in a long-run perspective.
Economic Policy, 28(73):45–100.
Bratti, M. and Leombruni, R. (2014). Local human capital externalities and wages at the firm
level: Evidence from Italian manufacturing. Economics of Education Review, 41(C):161–
175.
Brizzi, G.P. Del Negro, P. and Romano, A. e. (2007). Storia delle Universita in Italia. CISUI.
Brueckner, J. K. (2003). Strategic interaction among governments: An overview of empirical
studies. International Regional Science Review, 26(2):175–188.
Ciccone, A. and Peri, G. (2006). Identifying Human-Capital Externalities: Theory with
Applications. Review of Economic Studies, 73(2):381–412.
Ciccone, A. and Peri, G. (2011). Schooling Supply and the Structure of Production: Evidence
from US States 1950-1990. NBER Working Papers 17683, National Bureau of Economic
Research, Inc.
Clark, G. (2005). The condition of the working class in england, 1209-2004. Journal of
Political Economy, 113(6):1307–1340.
Davies, R. and Vadlamannati, K. C. (2013). A race to the bottom in labor standards? an
empirical investigation. Journal of Development Economics, 103(C):1–14.
De Giorgi, G., Pellizzari, M., and Redaelli, S. (2010). Identification of social interactions
through partially overlapping peer groups. American Economic Journal: Applied Eco-
nomics, 2(2):241–75.
De Mauro, T. (1970). Storia linguistica dell’Italia unita. Laterza, Bari.
Ertur, C. and Koch, W. (2007). Growth, technological interdependence and spatial exter-
nalities: theory and evidence. Journal of Applied Econometrics, 22(6):1033–1062.
40
Eurydice (1999). Two decades of reform in higher education in europe: 1980 onwards.
Eurydice studies, Eurydice.
Eurydice (2008). Higher education governance in europe: policies, structures, funding and
academic staff. Eurydice thematic studies.
Fabrizzi, F. (2008a). La provincia: storia istituzionale dell’ente locale pi discusso. dalla
riforma di crispi all’assemblea costituente. Technical Report 12/2008, federalismi.it.
Fabrizzi, F. (2008b). La provincia: storia istituzionale dell’ente locale pi discusso.
dall’assemblea costituente ad oggi. Technical Report 23/2008, federalismi.it.
Fabrizzi, F. (2008c). La provincia: storia istituzionale dell’ente locale pi discusso.
dall’unificazione amministrativa alla caduta della destra storica. Technical Report
7/2008, federalismi.it.
Fois, G. (1991). L’Universita di Sassari nell’Italia liberale. Universita di Sassari.
Galor, O. (2005). Unified growth theory. Development and comp systems, EconWPA.
Gennaioli, N., Porta, R. L., de Silanes, F. L., and Shleifer, A. (2013). Human Capital and
Regional Development. The Quarterly Journal of Economics, 128(1):105–164.
Hoxby, C. (2016). The dramatic economics of the u.s. market for higher education. The 8th
annual martin feldstein lecture, National Bureau of Economic Research.
Hoxby, C. and Turner, S. (2014). What high-achieving low-income students know about
college. American Economic Review, American Economic Association, 105(5):514–517.
Madsen, J. B. and Murtin, F. (2017). British economic growth since 1270: the role of
education. Journal of Economic Growth, 22(3):229–272.
OECD (1999). The Response of Higher Education Institutions to Regional Needs. OECD
Paris.
OECD (2008). OECD Review of Tertiary Education. Final Report, OECD, Paris. OECD
Paris.
Palombelli, G. (2012). L’evoluzione delle circoscrizioni provinciali dall’Unit d’Italia ad oggi.
I. Nicotra (eds.) Le Province dall’Unit alla Repubblica delle Autonomie. Donzelli editore,
Roma.
41
Parchet, R. (2014). Are local tax rates strategic complements or strategic substitutes? IdEP
Economic Papers 1407, USI Universit della Svizzera italiana.
Petracchi, A. (1962). Le Origini dell’ Ordinamento comunale e provinciale italiano: sto-
ria della legislazione piemontese sugli enti locali dalla fine dell’antico regime al chi-
udersi dell’ eta cavouriana (1770-1861). Pubblicazioni dell’Istituto per la scienza
dell’amministrazione pubblica. N. Pozza.
Pini, A. I. (2000). Le ”nationes” studentesche nel modello universitario bolognese nel medio
evo. Romano, Brizzi (a cura di), pp. 21-30. Bologna, CLUEB.
Schtz, G., Ursprung, H., and Woessmann, L. (2005). Education policy and equality of
opportunity. CESifo Working Paper Series 1518, CESifo Group Munich.
Silvestri, A. (2006). La nascita delle Facolta di Ingegneria e di Architettura in Italia. Cuzzolin
Editore.
Smith, W. (2010). Public Universities and the Democratization of the Core Public Sphere.
Public Universities and the Public Sphere. Palgrave Macmillan, New York.
Squicciarini, M. P. and Voigtlander, N. (2015). Human Capital and Industrialization:
Evidence from the Age of Enlightenment. The Quarterly Journal of Economics,
130(4):1825–1883.
Unioncamere (2011). Italia 150. le radici del futuro. il sistema camerale tra imprese e isti-
tuzioni a 150 anni dall’unita’ d’italia. Retecamere scrl - edizioni camere di commercio
ditalia, Unioncamere.
Valero, A. and Van Reenen, J. (2016). The economic impact of universities: Evidence from
across the globe. NBER Working Papers 22501, National Bureau of Economic Research,
Inc.
Woesmann, L. (2003). Schooling resources, educational institutions and student performance:
the international evidence. Oxford Bulletin of Economics and Statistics, 65(2):117–170.
42
Appendix A Data Appendix
Appendix A.1 HIU Register data
The register dataset on the History of Italian Universities (HIU) contains detailed and com-
plete information on institutions providing higher education in Italy, disaggregated at the
faculty level over the period 1861-2010. The register includes the following information:
• University name.
• Faculty name.
• 15 faculty field identifiers (Agricultural studies, Architecture, Chemical and Pharma-
ceutical studies, Economics and Statistics, Physical Education, Geo-Biological studies,
Law, Engineering, Educational studies, Litery studies and Philosophy, Foreign Lan-
guages, Medicine, Political Sciences, Psycology, and Mathematical Sciences), which we
aggregated into 7 teaching areas (Socio-Economic area, Physical Education, Law, En-
gineering and Architecture, Medicine, Sciences, and Humanities.) and 3 macro-areas
of science (Social Sciences, STEM, Science, Technology, Engineering and Mathematics
- STEM, and Humanities), according to the classification used by the Italian National
Statistical Office.
• Year of establishment of the faculty. This is recorded as the year when the faculty
is formally established as a provider of a higher education degree. Alongside with
this basic information, we recorded other potentially important ancillary dates i.e.
whether the faculty was built upon a pre-existing major of studies (e.g. belonging
to an existing faculty), the year the faculty was formally recognised as a provider of
University education, the date(s) when the faculty became part of a different university.
Details over institutional developments that motivate these alternative definitions are
available in the on-line appendix.
• Address of the university.
• Address of the faculty.
The final version of the register includes 582 faculties and (in) 78 universities registered
on the Italian territory at some point between 1861 and 2010.
About the 99% of faculties in our sample deliver standard BSc education. There are 5
faculties specialised in post-graduate education only (i.e. belonging to Universita’ Normale di
43
Pisa, Universita’ Europea di Roma, Universita’ di Scienze Gastronomiche, and IMT Lucca),
and 3 faculties that enroll foreign students only (i.e. belonging to Universita’ per Stranieri
di Siena, Perugia, and Reggio Calabria).
While they track very precisely the creation of new HE institutions during Italian his-
tory, as well as their change in status and governance, our data record only seven cases of
effective faculty closures. Four engineering faculties were closed due the re-organisation of
the Politechnic School of Milan, which in 2000 closed down its campuses in Como and Lecco
and opened brand new faculties in the Milan area. The faculty of Chemical studies was shut
down by the Ca Foscari University of Venice, in 1990 as well as the faculties of Environmental
Sciences and Mathematical Sciences in Urbino in 2006. This may underestimate the actual
closure HE education institutions. As a matter of fact, our sources do not allow to map
closures as precisely as start-ups. However, none of our original sources mentions significant
waves of closure of HE institutions during Italian history. (see on-line appendix and Brizzi
and Romano (2007) Vol.3 for details.).
Our data do not record changes in the exact address of each university and each faculty
at each point in time i.e. do not record changes of address over time. The university address
is identified by the address of the university dean, and the faculty address is the address of
the faculty dean. Both are collected as for 2010.
Appendix A.2 Neighborhood and distance matrices
We constructed two alternative local interaction matrices. The first one (1) is a simple conti-
guity matrix available from ISTAT. The second one (2) is a distance matrix, which contains
information on (a) linear distance, (b) travel distance, and (c) travel time. We computed
distances using the Google Maps API Geometry Library, and Google Maps Distance Matrix
API, respectively.
1 Contiguity matrix. Each province i is matched to (the HIU indicators of) its neighbors,
which we define as provinces j that share a border with province i. Notice that due to
the process of provinces creation discussed above, the neighbors of each province i in
most cases change over time. In practice, consider the example of province k, which
is a neighbor of province i at time t. Imagine that at time t + 1 a new province z is
established, which covers the geographical territory of k that shares a border with i.
Our data register this changes overtime so that until time t province k is recorded as
being neighbor of province i, while starting from time t+ 1 k is no longer adiacent to
i, while province z appears in the data, which shares a border with i.
44
2a Linear distance matrix. Each province i is matched to (the HIU indicators of) all other
provinces j 6= i, and for each (i, j) couple, the linear distance is recorded between i
and j. This is the distance in Km calculated “on-air”, by drawing, and measuring the
lenght of a straight line between the capitals of province i and province j (capoluoghi
di provincia). Notice that, as long as the capitals of provinces do not change over
time, the distance between provinces remains constant over our sample. In the case of
provinces with multiple “capoluoghi”, the largest capoluogo is considered as capital.
However there are only two cases in Italy (Barletta-Andria-Trani, and Pesaro-Urbino).
2b Travel distance matrix. Each province i is matched to (the HIU indicators of) all other
provinces j 6= i, and for each (i, j) couple, the travel distance is recorded between i and
j. This is the lowest distance in Km that one needs to travel from/to i to/from province
j, with whatever transport available (plane, car, train). Notice that these distances are
recorded in 2016, so they take a cross-sectional picture of travel connections between
italian provinces in that specific year i.e. they do not take into account the process of
infrastructures building that has occurred during italian history. The same convention
as in the case of linear distances applies to Barletta-Andria-Trani, and Pesaro-Urbino.
2c Travel time matrix. Each province i is matched to (the HIU indicators of) all other
provinces j 6= i, and for each (i, j) couple, the travel time is recorded between i and j.
This is the lowest time in minutes, that one needs to travel from/to i to/from province
j, with whatever transport available (plane, car, train). Also travel time is recorded
in 2016, so it does not take into account the process of infrastructures building and
innovation that has occurred during italian history. The usual convention as in the
case of distances applies to Barletta-Andria-Trani, and Pesaro-Urbino.
Appendix A.3 Additional province level variables
We collected historical indicators on province level economic and social characteristics from
Unioncamere (2011). Historical series were completed using the direct Census sources.
Total population: Population of residents in the province, all ages (Source: Union-
camere, 2011).
Total population in the 0-14 cohort: Population of residents aged 0-14 in the province
(Source: Italian Census data, 1951-2010).
Provincial value added per capita (worker): i.e. total provincial value added divided
the total population (active population) of the province (Source: Unioncamere, 2011).
45
Original historical series expressed in nominal terms (in lira, current values). Real
figures were obtained by applying the VA deflator at constant 1911 prices, available
from ISTAT.
Provincial participation rates: Active population, as a percentage of total population
(Source: Unioncamere, 2011).
Share of active population in agriculture industry services: number of workers in agri-
culture industry services as a share of total workers (Source: Unioncamere, 2011).
Census data covers the period 1861-2010. The collection years are 1861, 1871, 1891, 1901,
1911, 1921, 1931, 1936, 1951, 1961, 1971, 1981, 1991, 2001, 2010. Census data are not
available for 1881, so we retrieved them by linear interpolation.
Appendix B Additional Tables
Table B-1: Number of Faculties in Italy by field of study: 1870 and 2010
Faculty 1870 2010
1. HUMANITIES 16 127Education 1 34Languages 1 24Literature 12 54Psychology 2 15
2. SCIENTIFIC AND MEDICAL ST. 51 186Agriculture 9 37Chemistry&Pharmacy 18 31Geology&Biology 1 3Scientific studies 15 48Architecture 3 23Engineering 5 44
3. SOCIAL SCIENCES 42 210Medical studies 19 39Economics&Statistics 1 68Law 21 55Socio-political studies 1 48
Notes: There are five faculty fields that first appeared in Italian universities after1870 that are Education (1876), Foreign Languages (1954), Geology and Biology(1993) and Psychology (1971).
46
Tab
leB
-2:
Om
itte
dnei
ghb
ourh
ood
vari
able
san
d2S
LS:
full
set
ofco
effici
ents
[1]
base
lin
e[3
]p
re-u
nit
ari
an
[4]
pre
-un
itari
an
sam
ple
pro
vin
ces
pro
v.
(lagged
10
y.)
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
OL
SF
E2SL
SF
EP
an
el
A)
avera
ge
neig
hb
ou
rap
pro
ach
tota
ln
o.fa
cult
ies
in−i
–0.4
8***
–1.1
4***
–0.0
9–1.0
8**
–0.1
1–0.9
4*
(0.1
4)
(0.3
9)
(0.0
8)
(0.4
9)
(0.1
0)
(0.5
0)
no.
ofu
niv
ersi
ties
1.4
8***
1.3
6**
1.8
2***
2.0
0***
1.6
2**
2.1
1***
(0.5
5)
(0.5
7)
(0.6
4)
(0.6
3)
(0.6
1)
(0.5
5)
no.
ofA
-lev
elu
niv
ersi
ties
1.0
6***
0.8
6**
0.8
7**
0.3
10.9
5***
0.5
9(0
.34)
(0.3
4)
(0.3
4)
(0.4
1)
(0.3
1)
(0.3
9)
no.
ofp
riva
teu
niv
ersi
ties
2.3
8***
2.2
0***
1.9
3***
1.4
1**
2.2
8***
1.7
1**
(0.5
7)
(0.5
6)
(0.5
3)
(0.5
7)
(0.7
1)
(0.6
5)
Pan
el
B)
pair
wis
eap
pro
ach
tota
ln
o.fa
cult
ies
inj
–0.0
8***
–0.1
6***
–0.0
4**
–0.1
3**
–0.0
5*
–0.1
2***
(0.0
3)
(0.0
5)
(0.0
2)
(0.0
6)
(0.0
2)
(0.0
4)
no.
ofu
niv
ersi
ties
2.1
7***
2.2
0***
2.3
4***
2.4
8***
2.1
7***
2.3
7***
(0.5
2)
(0.4
9)
(0.5
6)
(0.4
9)
(0.5
4)
(0.4
5)
no.
ofA
-lev
elu
niv
ersi
ties
0.6
5**
0.6
4***
0.4
8*
0.4
6**
0.5
6**
0.5
7**
(0.2
7)
(0.2
4)
(0.2
6)
(0.2
1)
(0.2
6)
(0.2
2)
no.
ofp
riva
teu
niv
ersi
ties
1.9
5***
1.7
9***
1.4
7***
1.2
9***
1.8
2***
1.7
7***
(0.5
4)
(0.5
0)
(0.4
5)
(0.3
8)
(0.5
9)
(0.5
4)
Note
s:.
Fu
llse
tof
coeffi
cien
tsre
fers
toO
LS
FE
an
d2S
LS
FE
esti
mate
sre
port
edin
Table
6.
All
spec
ifica
tion
sin
clu
de
pro
vin
cefi
xed
effec
ts(P
an
elA
),p
rovin
cial
pair
fixed
effec
ts(P
an
elB
),an
dre
gion
-by-y
ear
fixed
effec
ts.
In2S
LS
esti
mate
s,th
ein
stru
men
tsare
inte
ract
ion
sof
init
ial
con
dit
ion
sw
ith
hig
her
edu
cati
on
refo
rms,
as
rep
ort
edin
tab
leB
-7.
Sta
nd
ard
erro
rscl
ust
ered
at
the
pro
vin
cele
vel.
Sig
nifi
can
cele
vels
:∗
:10%
∗∗:
5%
∗∗∗:
1%
.
47
Table B-3: Competition across and within field of study: first stage
humanities (HH) stem (ST) social sciences (SS)(ICj in HH)*(L. 2102/1923) 0.34*** –0.18*** 0.09 1.05***
(0.11) (0.06) (0.08) (0.08)(ICj in HH)*(L. 1592/1933) 0.08 –0.23*** 0.26*** 0.41***
(0.06) (0.06) (0.07) (0.07)(ICj in HH)*(L. 910/1969) 0.16** –0.22** 0.49*** 0.06
(0.07) (0.10) (0.04) (0.11)(ICj in HH)*(L. 766/1973) 0.04 –0.12** 0.12*** 0.30***
(0.03) (0.05) (0.04) (0.10)(ICj in HH)*(L. 382/1980) 0.01 0.03 –0.05 0.01
(0.02) (0.02) (0.04) (0.05)(ICj in HH)*(L. 168/1989) –0.05** 0.06** –0.34*** 0.11**
(0.02) (0.03) (0.06) (0.04)(ICj in HH)*(L. 245-341/1990) 0.41*** 0.40*** –0.53*** 0.14**
(0.11) (0.09) (0.08) (0.07)(ICj in HH)*(L. 59/1997) 0.58*** 0.36*** –0.26*** 0.13
(0.14) (0.11) (0.08) (0.13)(ICj in ST)*(L. 2102/1923) 0.14*** 0.04 0.15*** –0.23***
(0.02) (0.04) (0.04) (0.04)(ICj in ST)*(L. 1592/1933) –0.03 –0.10** 0.15*** 0.09**
(0.02) (0.05) (0.02) (0.04)(ICj in ST)*(L. 910/1969) –0.05 –0.18*** 0.22*** 0.15***
(0.04) (0.03) (0.02) (0.04)(ICj in ST)*(L. 766/1973) –0.12*** –0.04* 0.12** 0.23***
(0.03) (0.02) (0.05) (0.07)(ICj in ST)*(L. 382/1980) –0.01 –0.04 0.00 0.04**
(0.01) (0.02) (0.02) (0.02)(ICj in ST)*(L. 168/1989) –0.02 0.05** 0.02 0.00
(0.01) (0.02) (0.02) (0.02)(ICj in ST)*(L. 245-341/1990) 0.11*** –0.02 0.05 –0.11***
(0.03) (0.04) (0.03) (0.03)(ICj in ST)*(L. 59/1997) 0.17*** –0.03 0.07 –0.12**
(0.04) (0.03) (0.05) (0.06)(ICj in SS)*(L. 2102/1923) –0.17*** 0.10*** 0.02 0.11***
(0.03) (0.03) (0.03) (0.04)(ICj in SS)*(L. 1592/1933) 0.09*** 0.11*** 0.10** –0.09**
(0.03) (0.03) (0.04) (0.04)(ICj in SS)*(L. 910/1969) 0.18*** 0.09*** 0.03 –0.00
(0.06) (0.02) (0.04) (0.06)(ICj in SS)*(L. 766/1973) 0.17*** 0.02 0.01 –0.29***
(0.03) (0.01) (0.03) (0.07)(ICj in SS)*(L. 382/1980) –0.00 0.03 0.08*** –0.06**
(0.01) (0.02) (0.02) (0.03)(ICj in SS)*(L. 168/1989) –0.06*** 0.09*** 0.14*** –0.01
(0.02) (0.03) (0.04) (0.02)(ICj in SS)*(L. 245-341/1990) –0.22*** 0.16*** 0.29*** 0.09**
(0.05) (0.03) (0.06) (0.04)(ICj in SS)*(L. 59/1997) –0.15*** –0.00 0.04 0.09
(0.05) (0.02) (0.04) (0.07)R sq. 0.82 0.89 0.96 0.97 0.87 0.93N 32644 32644 32644 32644 32644 32644
Notes:. First stage of IV FE estimates reported in Table 7. All specifications include provincial pairfixed effects, region-by-year fixed effects, and the usual set of provincial controls. Standard errorsclustered by province are reported in parentheses. Significance levels: ∗ : 10% ∗∗ : 5% ∗∗∗:1%.
48
Table B-4: HE supply Competition: local demand effects
[1] [2] [3]OLS FE 2SLS FE Obs.
Panel A) baseline specification 34912total no. faculties in j –0.05* –0.12***
(0.02) (0.04)no. of universities 2.22*** 2.42***
(0.54) (0.45)no. of private universities 1.78*** 1.72***
(0.59) (0.54)no. of elite universities 0.55** 0.56**
(0.27) (0.23)Panel B) drop metropolitan cities 22814tot. no. faculties in j –0.11** –0.22***
(0.04) (0.07)no. of universities 1.93*** 2.14***
(0.50) (0.40)no. of private universities 0.27 0.16
(0.52) (0.43)no. of elite universities 0.77*** 0.71***
(0.27) (0.23)Panel C) control for population size 34912total no. faculties in j –0.04 –0.12***
(0.02) (0.04)no. of universities 2.01*** 2.24***
(0.60) (0.50)no. of private universities 1.68*** 1.74***
(0.60) (0.55)no. of elite universities 0.47* 0.54**
(0.27) (0.23)total population (log) 1.37* 0.75
(0.76) (0.68)Panel D) control for share of active in industry sector 34912total no. faculties in j –0.05* –0.12***
(0.02) (0.04)no. of universities 2.17*** 2.42***
(0.55) (0.45)no. of private universities 1.82*** 1.72***
(0.58) (0.53)no. of elite universities 0.55** 0.55**
(0.27) (0.23)share of active in the industry sector –0.01 –0.01
(0.01) (0.01)participation rate 0.01 0.01
(0.01) (0.01)Panel E) control for share of 0-14 years old 11442total no. faculties in j –0.05* –0.09*
(0.03) (0.05)no. of universities 1.92*** 2.17***
(0.65) (0.59)no. of private universities 1.96*** 1.78***
(0.47) (0.43)0-15 cohort size 0.03 0.05
(0.08) (0.08)Panel F) placebo: “alphabetical” neighbors 48337total no. faculties in j 0.009 0.063
(0.020) (0.061)no. of universities 1.750*** 1.796***
(0.587) (0.554)no. of elite universities 0.986*** 1.055***
(0.284) (0.261)no. of private universities 2.421*** 2.224***
(0.626) (0.600)
Notes: Full set of coefficients refers to OLS FE and 2SLS FE estimatesreported in Table 8. All specifications include provincial pair fixed effects,and region-by-year fixed effects. In 2SLS estimates, the instruments areinteractions of initial conditions with higher education reforms. Standarderrors clustered at the province level. Significance levels: ∗ : 10% ∗∗: 5% ∗ ∗ ∗: 1%.
49
Tab
leB
-5:
Com
pet
itio
nin
HE
supply
:dis
tance
and
tim
em
atri
ces
Pan
el
A)
travel
dis
tan
ce
(Km
)[1
]w
ith
in90
Km
[2]
bet
wee
n90
an
d180
Km
[3]
bet
wee
n180
an
d270
Km
[5]
bet
wee
n270
an
d360
Km
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
tota
ln
o.fa
cult
ies
inj
–0.0
46**
–0.1
36***
–0.0
19
0.0
21
–0.0
16
–0.0
24*
0.0
09
–0.0
02
(0.0
18)
(0.0
44)
(0.0
14)
(0.0
14)
(0.0
12)
(0.0
13)
(0.0
11)
(0.0
07)
Ob
serv
atio
ns
4006
940
069
91090
91090
100647
100647
78450
78450
K-P
rkW
ald
F-s
tat
37.3
98151.8
15
155.4
45
75.2
94
K-P
rkL
M-s
tat
42.0
4052.9
40
49.9
15
55.7
50
p-v
alu
e0.
000
0.0
00
0.0
00
0.0
00
Han
sen
J-s
tat
9.52
58.1
64
6.2
71
6.8
67
p-v
alu
e0.
217
0.3
18
0.5
08
0.4
43
Pan
el
B)
travel
tim
e(m
inu
tes)
[1]
wit
hin
80m
inu
tes
[2]
bet
wee
n80
an
d160
min
.[3
]b
etw
een
160
an
d240
min
.[5
]b
etw
een
240
an
d360
min
.O
LS
FE
2SL
SF
EO
LS
FE
2S
LS
FE
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
tota
ln
o.fa
cult
ies
inj
–0.0
90**
*–0
.162
***
–0.0
14
–0.0
03
–0.0
03
–0.0
18
0.0
17
0.0
07
(0.0
28)
(0.0
45)
(0.0
13)
(0.0
14)
(0.0
08)
(0.0
12)
(0.0
11)
(0.0
07)
Ob
serv
atio
ns
2940
329
403
102981
102981
115585
115585
84469
84469
K-P
rkW
ald
F-s
tat
28.5
82193.5
43
178.6
43
73.5
52
K-P
rkL
M-s
tat
38.7
5353.8
91
52.4
01
53.7
02
p-v
alu
e0.
000
0.0
00
0.0
00
0.0
00
Han
sen
J-s
tat
9.27
09.7
24
5.5
20
6.0
45
p-v
alu
e0.
234
0.2
05
0.5
97
0.5
34
Note
s:.
Sp
ecifi
cati
onas
inT
able
6co
lum
n[4
]w
ith
pre
-un
itari
an
pro
vin
ces
du
rin
gth
een
tire
per
iod
an
d10
years
lags
of
regre
ssor
an
dco
ntr
ols
.In
2SL
Ses
tim
ates
,th
eto
tal
no.
offa
cult
ies
inpro
vin
cej
isin
stru
men
ted
by
the
init
ial
con
dit
ions
(i.e
.nu
mb
erof
ma
jors
inj
in1861)
inte
ract
edby
ab
atte
ryof
du
mm
ies
for
hig
her
edu
cati
onre
form
sin
Italy
.S
eeT
ab
leB
-7b
elow
for
det
ail
s.A
llsp
ecifi
cati
on
sin
clu
de
pro
vin
cial
pair
fixed
effec
ts,
regi
on-b
y-y
ear
fixed
effec
ts,
and
the
usu
al
set
of
pro
vin
cial
contr
ols
.S
tan
dard
erro
rscl
ust
ered
by
pro
vin
ceare
rep
ort
edin
pare
nth
eses
.S
ign
ifica
nce
leve
ls:
∗:
10%
∗∗:
5%∗∗∗:
1%
.
50
Tab
leB
-6:
HE
supply
Com
pet
itio
n:
het
erog
enei
ty(fi
rst
stag
es)
[1]
pre
-war
[2]
post
-war
[3]
Nort
h[4
]C
entr
e-S
ou
th[5
]In
tra-r
egio
nal
[6]
Inte
r-re
gio
nal
(IC
j)*
(L.
2102
/192
3)0.
15**
*–
0.2
1***
0.1
1**
0.1
3**
0.2
0***
(0.0
3)–
(0.0
4)
(0.0
5)
(0.0
5)
(0.0
7)
(IC
j)*
(L.
1592
/193
3)–
–0.1
4***
0.2
0***
0.2
3***
0.0
6–
–(0
.03)
(0.0
6)
(0.0
3)
(0.0
7)
(IC
j)*
(L.
910/
1969
)–
0.1
9***
0.1
8***
0.2
8***
0.2
3***
0.2
2***
–(0
.02)
(0.0
2)
(0.0
3)
(0.0
2)
(0.0
4)
(IC
j)*
(L.
766/
1973
)–
–0.0
0–0.0
30.0
2–0.0
5*
0.0
8*
–(0
.03)
(0.0
3)
(0.0
6)
(0.0
3)
(0.0
4)
(IC
j)*
(L.
382/
1980
)–
0.1
9***
0.2
0***
0.1
8**
0.1
7***
0.1
7**
–(0
.04)
(0.0
4)
(0.0
7)
(0.0
4)
(0.0
7)
(IC
j)*
(L.
168/
1989
)–
0.1
9***
0.1
6***
0.2
3**
0.2
3***
0.0
8–
(0.0
5)
(0.0
6)
(0.0
9)
(0.0
7)
(0.0
6)
(IC
j)*
(L.
245/
1990
)–
–0.0
2–0.0
5**
0.0
2–0.0
2–0.0
5–
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
4)
(IC
j)*
(L.
59/1
997)
––0.0
00.0
3–0.0
4–0.0
10.0
1–
(0.0
3)
(0.0
3)
(0.0
3)
(0.0
2)
(0.0
6)
Ob
serv
atio
ns
1728
314330
17785
17275
20234
14687
Note
s:F
irst
stag
eof
IVF
Ees
tim
ate
sre
port
edin
Tab
le??.
All
spec
ifica
tion
sin
clu
de
pro
vin
cialp
air
fixed
effec
ts,re
gio
n-b
y-
year
fixed
effec
ts,
and
the
usu
alse
tof
pro
vin
cialco
ntr
ols
.Sta
nd
ard
erro
rscl
ust
ered
by
pro
vin
ceare
rep
ort
edin
pare
nth
eses
.S
ign
ifica
nce
level
s:∗
:10
%∗∗
:5%
∗∗∗:
1%
.
51
Table B-7: Competition in HE supply: alternative definitions of pre-unitarian provinces: 1ststage
[1] [2] [3] [4](ICj)*(L. 2102/1923) 0.16*** 0.17*** 0.19*** 0.13***
(0.03) (0.03) (0.03) (0.04)(ICj)*(L. 1592/1933) 0.09*** 0.10*** 0.14*** 0.11***
(0.03) (0.02) (0.02) (0.03)(ICj)*(L. 910/1969) 0.21*** 0.21*** 0.24*** 0.20***
(0.02) (0.02) (0.02) (0.02)(ICj)*(L. 766/1973) 0.05*** 0.06*** 0.05*** 0.05***
(0.02) (0.02) (0.02) (0.02)(ICj)*(L. 382/1980) 0.11** 0.04* 0.04*** 0.06**
(0.04) (0.02) (0.02) (0.02)(ICj)*(L. 168/1989) 0.06* 0.04** 0.04*** 0.04***
(0.03) (0.02) (0.01) (0.01)(ICj)*(L. 245-341/1990) 0.17*** 0.16*** 0.20*** 0.22***
(0.05) (0.04) (0.04) (0.05)(ICj)*(L. 59/1997) 0.16* 0.17*** 0.20*** 0.20***
(0.08) (0.05) (0.02) (0.03)R sq. 0.94 0.95 0.96 0.97Observations 42813 42813 68410 35328
Notes: First stage of IV FE estimates reported in Table 11. Allspecifications include provincial pair fixed effects, region-by-year fixedeffects, and the usual set of provincial controls. Standard errors clus-tered by province are reported in parentheses. Significance levels:∗ : 10% ∗∗ : 5% ∗ ∗ ∗: 1%.
52
Tab
leB
-8:
Sen
siti
vit
yan
alysi
s:al
tern
ativ
eem
pir
ical
stra
tegi
es
[1]
2S
LS
FE
[2]
2S
LS
FE
[3]
2S
LS
FE
[4]
2S
LS
FE
[5]
2S
LS
FE
[6]
2S
LS
FE
no.
facu
ltie
sin
2nd
deg
ree
nei
ghb
our
–0.0
1*
(0.0
0)
(IC
j)*
(L.
2102
/192
3)0.1
9***
0.2
7***
0.1
5***
0.2
6***
0.1
3***
(0.0
5)
(0.0
4)
(0.0
4)
(0.0
4)
(0.0
4)
(IC
j)*
(L.
1592
/193
3)0.1
4***
0.1
6***
0.1
6***
0.1
7***
0.1
1***
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
3)
(IC
j)*
(L.
910/
1969
)0.2
2***
0.1
0***
0.2
3***
0.2
4***
0.1
9***
0.2
0***
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
2)
(IC
j)*
(L.
766/
1973
)0.0
6***
0.0
7***
0.0
8***
0.0
6***
0.0
10.0
5***
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
2)
(0.0
3)
(0.0
2)
(IC
j)*
(L.
382/
1980
)–0.0
3**
0.0
00.0
10.1
7***
0.0
6**
(0.0
1)
(0.0
1)
(0.0
2)
(0.0
4)
(0.0
2)
(IC
j)*
(L.
168/
1989
)0.0
2*
0.0
10.0
3***
0.1
7***
0.0
4***
(0.0
1)
(0.0
1)
(0.0
1)
(0.0
5)
(0.0
1)
(IC
j)*
(L.
245/
1990
)0.1
0***
0.1
1***
0.1
3***
0.2
2***
(0.0
2)
(0.0
3)
(0.0
3)
(0.0
5)
(IC
j)*
(L.
59/1
997)
0.0
8***
0.1
0***
0.1
4***
0.2
0***
(0.0
3)
(0.0
3)
(0.0
3)
(0.0
3)
Rsq
.0.9
30.9
90.9
40.9
60.9
30.9
7O
bse
rvat
ion
s109817
2653
34904
35060
32210
35328
Note
s:.
Fir
stst
age
resu
lts
ofre
gres
sion
sre
port
edin
Tab
le12.
All
spec
ifica
tion
sin
clu
de
pro
vin
cial
pair
fixed
effec
ts,
regio
n-b
y-
year
fixed
effec
ts,
and
the
usu
alse
tof
pro
vin
cialco
ntr
ols
.D
etail
sare
Tab
le12.
Sta
nd
ard
erro
rscl
ust
ered
by
pro
vin
ceare
rep
ort
edin
par
enth
eses
.S
ign
ifica
nce
level
s:∗
:10%
∗∗:
5%
∗∗∗:
1%
.
53
Tab
leB
-9:
Val
ue
ofH
Esu
pply
:F
irst
stag
e
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
OL
SF
E2S
LS
FE
IC
i∗R
21/30
0.28
19**
*0.
2816***
0.2
988***
0.2
986***
0.2
988***
0.2
986***
0.0
593
0.0
610
(0.0
652)
(0.0
650)
(0.0
732)
(0.0
719)
(0.0
732)
(0.0
719)
(0.0
499)
(0.0
503)
IC
i∗R
31/36
0.18
42**
*0.
1844***
0.2
035***
0.1
931***
0.2
035***
0.1
931***
–0.0
971
–0.1
001
(0.0
548)
(0.0
546)
(0.0
545)
(0.0
537)
(0.0
545)
(0.0
537)
(0.0
956)
(0.0
924)
IC
i∗R
61/70
0.15
55**
*0.
2758***
0.2
742***
0.2
861***
0.2
742***
0.2
861***
–0.0
116
–0.0
036
(0.0
427)
(0.0
921)
(0.0
997)
(0.0
976)
(0.0
997)
(0.0
976)
(0.0
772)
(0.0
757)
IC
i∗R
71/80
0.02
77(0
.042
5)IC
i∗R
81/90
0.20
47(0
.134
0)IC
i∗R
91/00
–0.0
096
(0.0
393)
Ft−
20
1.1
100***
1.1
095***
(0.1
348)
(0.1
337)
Ob
serv
atio
ns
937
937
937
937
937
937
874
874
Note
s:.
Fir
stst
age
ofIV
FE
esti
mate
sre
port
edin
Tab
le13.
All
spec
ifica
tion
sin
clu
de
pro
vin
cefi
xed
effec
tsan
dre
gion
-by-y
ear
fixed
effec
ts,
pop
ula
tion
gro
wth
,p
art
icip
ati
on
rate
an
dsi
zeof
the
ind
ust
ryse
ctor.
Sta
nd
ard
erro
rscl
ust
ered
by
pro
vin
cear
ere
por
ted
inp
are
nth
eses
.S
ign
ifica
nce
leve
ls:
∗:
10%
∗∗:
5%
∗∗∗:
1%
.
54
Recommended